Shifts in U.S. corruption enforcement priorities, tactics and appetites dominated headlines in 2025, but they were not the only signs of adaptation. The U.K. has seen its own developments, with further movement expected in 2026.

At the start of 2025, the U.K.’s SFO, along with France’s Parquet National Financier (PNF) and the Office of the Attorney General of Switzerland, launched a tripartite Anti-Corruption Task Force, followed by new, more detailed External Guidance on Corporate Co-Operation and Enforcement in relation to Corporate Criminal Offending (2025 Co-Operation Guidance). In November 2025, the SFO issued Guidance on Evaluating a Corporate Compliance Programme (GECCP) and, in December, the U.K. Home Office published the UK Anti-Corruption Strategy 2025 (Strategy). Then, in a surprise move in January 2026, Nick Ephgrave, head of the SFO, announced that he will be retiring about half-way through what was supposed to be a five-year term, leaving the future of SFO enforcement paused in a moment of uncertainty.

This first article in a two-part series examining U.K. enforcement developments analyzes the GECCP and what it means for companies. The second article will examine the Strategy and staffing changes at the SFO.

See our two-part series on the SFO’s Co‑Operation Guidance: “A Hard Sell for Self-Reporting” (May 21, 2025), and “Investigation Expectations” (Jun. 4, 2025).

Setting Procedural Expectations

The GECCP primarily provides guidance to companies about when they can expect to have their compliance programs evaluated. It is “an important publication for organizations,” Paul Nash, a managing director at Nardello & Co, told the Anti-Corruption Report.

The guidance lays out the six “scenarios” in which the SFO may need to analyze a compliance program. Such an analysis will take place:

1. to determine if a prosecution of the organization is in the public interest under the Joint SFO-CPS Corporate Prosecution Guidance;
2. to consider whether a DPA is appropriate under the Deferred Prosecution Agreements Code of Practice;
3. to include compliance terms and/or a monitorship as part of a DPA;
4. if an organization has a defense of “adequate procedures” to a charge of failure of a commercial organization to prevent bribery, under Section 7 of the U.K. Bribery Act 2010 (UKBA);
5. if an organization has a defense of “reasonable procedures” to a charge of failure of a commercial organization to prevent fraud, under s.199 of the Economic Crime and Corporate Transparency Act 2023 (ECCTA); or
6. if the existence and nature of the compliance program is a relevant factor for sentencing considerations.

The GECCP further explains in detail the role that a compliance program evaluation plays in each situation, providing a brief summary of the SFO’s expectations in the respective scenario.

For example, when determining whether to enter into a DPA, “the relevant evaluation is of the effectiveness and the proactive nature approach of the compliance programme, at the time of the offending, the time of reporting and at the time of entering into the DPA,” the GECCP summarizes. In contrast, when determining the terms of a DPA, “the relevant evaluation is to determine what changes to the compliance programme are fair, reasonable and proportionate and would result in a robust compliance programme, and what compliance improvements are fair, reasonable and proportionate to include in a monitorship agreement.”

The goal of the GECCP is to set expectation for corporations “so they know exactly when to expect requests” from the SFO about a compliance program, SFO Director of Operations Emma Luxton shared while speaking at the American Conference Institute’s 42^nd Annual Conference on FCPA in December 2025 (ACI Conference).

Thus, the GECCP is primarily procedural, Nash explained, aimed at “clarifying when the SFO will assess a compliance program, rather than setting out how compliance controls will be evaluated.”

The GECCP is the “first formal confirmation that companies should expect the SFO to interrogate their compliance programs at an early stage of an investigation, using its full armory of investigative powers,” Simon Airey, a partner at McDermott Will & Schulte, told the Anti-Corruption Report.

For more from the ACI Conference, see our three-part series “2025 in Review”: White-Collar Enforcement the “Right Way” Remains a Priority (Dec. 17, 2025), DOJ Perspectives on How the Blanche Memo Restarted FCPA Enforcement (Jan. 14, 2026); and “2025 in Review: Impact on In-House Teams and Their Defense Counsel” (Jan. 28, 2026).

A Renewed Focus on Compliance

Even as a purely procedural document, the GECCP serves as notice to corporations that the SFO will be looking at corporate compliance programs and taking those evaluations seriously.

The “biggest takeaway” from the GECCP is the “renewed focus by the SFO on corporate compliance,” Judith Seddon, a partner at Ashurst, told the Anti-Corruption Report. Against the backdrop of recent changes in how the U.K. handles white-collar crime, including the ECCTA, the renewed focus is “not surprising, not least because the SFO has made no secret of the fact that it would like to be at the forefront” of enforcing this new law, she said.

The GECCP is also a “clear indication of the direction that the SFO is moving in” with regard to enforcing the ECCTA, Audrey Koh, a partner at Pillsbury, told the Anti-Corruption Report. The SFO is looking for “risk-based, proportionate policies and procedures over tick-box checklists,” she observed.

Scant on Details

Despite the SFO’s enthusiasm for evaluating compliance programs as part of its enforcement efforts, the GECCP is scant on details about what a compliance program should look like. “As I was reading the GECCP, I kept looking for the punchline, but when I got to the end it seemed that it did not really say anything about the assessment itself,” Koh recalled.

Seddon agreed that “little in the [GECCP] is new,” noting that it just “reiterates previous messages about compliance being more than tick-box, and that any compliance program must be proportionate, risk-based and regularly reviewed.”

What Might Be Requested

While the majority of the GECCP simply recites the situations in which the SFO will consider a compliance program, a frequently asked questions (FAQ) section at the end of the document provides some small insights into how it will make those evaluations.

One of the three questions in the FAQ is “What sources of evidence will the SFO obtain to conduct the evaluation of a compliance programme?”

To make its evaluation, the SFO will need to look at a variety of documents related to a company’s compliance program, the FAQ explains. Additionally, it notes that the “sources of this information – in particular, sources of information concerning failures of a compliance programme – are also likely to be sources of information on wider questions such as direct or circumstantial evidence of criminality.”

The SFO’s information requests also are “likely to include documentary production orders relating to board minutes, and interviews with those ultimately responsible for compliance (i.e., the board),” Airey said.

A Determination on Effectiveness

The last question in the FAQ is telling as to what will be most relevant to the SFO in evaluating compliance programs. It asks, “What makes a compliance programme effective or not?”

The framing of this question clarifies that the SFO is “focusing on real world effectiveness of a company’s compliance program,” Koh said.

The answer to this question further illuminates how the SFO thinks about compliance in solid terms. “There are no set or preordained answers that entitle an organisation to (or disqualify it from) a specific result, decision or recommendation that its compliance programme is effective,” the GECCP says. “The SFO’s assessment will be a holistic one, based on the organisation’s individual circumstances,” it adds.

The GECCP specifically notes that having policies and procedures in place does not necessarily equate to effectiveness. “The SFO will seek to get behind the pronouncements and determine how policies and procedures translate into conduct on the ground,” it says.

On the flip side, the GECCP makes clear that a company can still be found to have an effective program even if it has had “isolated compliance failures.” To determine if a program is effective in such a situation, the SFO “will consider whether the compliance measures had sufficient systems and controls against circumvention.”

See “To Work Effectively, CCOs Need Authority, Autonomy and Information” (Nov. 6, 2024).

Compliance Evaluations for Purposes of “Failure to Prevent” Defenses

Of the six circumstances under which the SFO will review a corporate compliance program, two of the most notable focus on whether the company has a defense to allegations of failure to prevent fraud or bribery.

Failure to Prevent Bribery

It is a violation of the UKBA for a company to fail to prevent bribery, but, under Section 7 of the law, a company can avoid liability if it had in place “adequate procedures” to prevent such bribery. Associated guidance, first issued in 2013 by the Ministry of Justice, lays out six principles of bribery prevention procedures (UKBA Guidance), which the GECCP incorporates:

1. top level commitment;
2. risk assessment;
3. due diligence;
4. communication (including training); and
5. monitoring and review.

Failure to Prevent Fraud

The ECCTA came into full effect in September 2025. Like the UKBA, it requires the SFO to evaluate a company’s compliance program to determine whether it had “reasonable procedures” for the prevention of fraud. The U.K. Home Office issued guidance in November 2024 (ECCTA Guidance) which, like the UKBA Guidance, lays out six principles of a fraud prevention program that are incorporated into the GECCP.

“Rather than provide new insights, the [GECCP] reiterates and summarizes” the principles outlined in the ECCTA Guidance, Nash observed.

The GECCP extorts readers to “[n]ote . . . that an organisation could argue under ECCTA that it was not reasonable to have any procedures in place at all.”

“Under the ECCTA, an organization could argue that it was acceptable to not have any reasonable procedures in place at all,” Nash explained, which “contrasts with the [UKBA], where an organization requires adequate procedures in all circumstances.” However, it is unlikely that any large corporation subject to the ECCTA would take this position, he predicted.

The GECCP also notes that the ECCTA specifically calls for an assessment of a company’s procedures and not its “compliance programme,” more specifically.

Declining to Differentiate “Adequate” and “Reasonable” Procedures

The GECCP goes out of its way to highlight that the UKBA calls for an assessment of whether a company had “adequate” procedures in place, while the ECCTA calls for a review of whether the company had “reasonable” procedures. However, “there is limited guidance, or case law, of how to interpret ‘reasonable’ procedures and ‘adequate’ procedures, and the SFO has not provided further direction other than the guidance provided under the ECCTA and the Bribery Act,” Nash observed.

FAQ Does Not Differentiate

Although a question in the FAQ queries what the difference is “between ‘adequate’ or ‘reasonable’ procedures and an ‘effective compliance programme,’” the somewhat disappointing answer states that, beyond the UKBA and ECCTA Guidance, “there is no formal guidance or interpretation of what constitutes adequate or reasonable procedures (or an effective compliance programme).”

Notably, the GECCP does not mention the defense of reasonable procedures contained in the U.K. Criminal Finances Act 2017, Section 46, and explained in further detail in associated guidance. “That offence has yet to be successfully prosecuted and yet the SFO is a designated prosecutor for it,” Seddon observed.

A Distinction Without a Difference

The SFO’s reluctance to clearly delineate the standards of “adequate” and “reasonable” procedures is a signal to companies that this may be a distinction without a difference.

“I don’t think there is any functional difference between the two words,” Koh said.

“Etymologically, there is surely a difference between ‘adequate’ and ‘reasonable,’” Airey observed, “but this is not an issue that any sensible corporate will wish to be a test case for.”

“The functional difference [between the two words] is limited,” Sean Seelinger, a partner at Ropes & Gray, agreed. “Companies should focus on the substance of what they need to implement to mitigate risk and establish a defense under various jurisdictions whose laws they may be subject to rather than focusing on technical differences in terminology in U.K. legislation,” he told the Anti-Corruption Report.

Effectiveness Remains the Critical Measure

The SFO’s guidance on reasonable versus adequate procedures, such as it is, confirms that program effectiveness is really the critical analysis. “If there was ever any doubt that the adequate and reasonable procedure defenses represented different thresholds, [the GECCP] puts that issue to bed,” Seddon said.

“There used to be some debate as to whether an isolated compliance failure would render a program ‘inadequate,’” Seddon recounted. The GECCP “provides the answer, if one were needed,” she said.

“A core message from the GECCP is that the SFO is ultimately concerned with assessing whether the compliance program is effective in practice,” Airey observed. Based on the FAQ, the SFO’s evaluation will focus on program design, and substance over form, he said.

Look to the DOJ and AFA Guidance

Tucked into the end of the FAQ answer about “adequate” versus “reasonable” procedures is a curious bit of advice from the SFO. “References to external sources may assist the determination of what constitutes an effective compliance programme,” it suggests, and then directs companies “with a US link” to review the DOJ’s Evaluation of Corporate Compliance Programs (ECCP) and those “with a French link” to review the guidance issued by the French Anti-Corruption Agency (AFA and AFA Guidance).

Working Within Limitations

One reason that the SFO may have opted to reference the AFA Guidance and ECCP rather than issue its own, more detailed guidance is because it felt limited in its role.

The UKBA Guidance was issued by the Ministry of Justice and the ECCTA Guidance was issued by the Home Office, not the SFO, Airey noted. “Accordingly, the SFO may have felt constrained in its ability to expand on the contents of those documents,” he posited.

The ECCTA Guidance, in particular, is very new. “In view of the fact that it is only a few months since the ECCTA Guidance on reasonable procedures was published, after a broad consultation across industry, it does not come as a surprise that the SFO has chosen not to replicate that exercise,” Seddon said.

Additionally, the SFO is a “younger” enforcement agency than the DOJ and still needs judicial approval for all of its settlements. This means the SFO must be “more careful and not fetter any of its discretion,” Koh suggested.

More detailed guidance from the SFO may have also proven confusing given the UKBA Guidance, ECCTA Guidance and other sources of compliance guidance. “It can be argued that providing another set of guidelines would confuse, not clarify, what is expected from organizations when seeking to demonstrate ‘effectiveness,’” Nash suggested.

Alignment With International Standards

The reference to the ECCP and the AFA Guidance is also an indication that the SFO intends to align its compliance expectations with international standards.

“The SFO appears to have deliberately positioned [the GECCP] as procedural, signaling alignment with established international frameworks rather than creating a parallel U.K. standard,” Nash said. “For organizations operating internationally, this promotes consistency and reduces the risk of conflicting guidance.”

The SFO’s “willingness to cross-refer may indicate future SFO endorsement of these more granular frameworks for benchmarking purposes when stress-testing a compliance program, while preserving flexibility in U.K. enforcement,” Seelinger suggested.

The SFO’s reference to DOJ and AFA guidance was “quite clever,” Koh observed, as it “signals alignment with an international enforcement consensus without binding itself.” Doing so is “alignment by reference and quietly importing the substance of these other systems and guidance without codifying it,” she elaborated. “It’s quite subtle and it’s quite clever.”

See our two-part series on emerging global compliance standards: “DOJ, OECD and World Bank Guidance” (Oct. 22, 2025), and “AFA, SFO and Eight Common Compliance Elements” (Nov. 5, 2025).

The prosecution of Mexican national Ramon Alexandro Rovirosa Martinez on charges of conspiracy and aiding and abetting violations of the FCPA is one of the first enforcement actions after Deputy AG Todd Blanche issued new FCPA enforcement guidance in a June 2025 memorandum (Blanche Memo).

While the underlying facts amount to a “typical” bribery pattern, the case illustrates how the DOJ can meet its dual goals of protecting U.S. businesses and pursuing individual violators. The Anti-Corruption Report spoke with attorneys in both the U.S. and Mexico, including former prosecutors in both countries, to understand the implications of the verdict.

See our three-part series “2025 in Review”: White-Collar Enforcement the ‘Right Way’ Remains a Priority (Dec. 17, 2025), DOJ Perspectives on How the Blanche Memo Restarted FCPA Enforcement (Jan. 14, 2026), and Impact on In-House Teams and Their Defense Counsel (Jan. 28, 2026).

Bribery in Mexico

Defendants Ramon Alexandro Rovirosa Martinez (Rovirosa) and Mario Alberto Avila Lizarraga (Avila) (collectively, Defendants) were both Mexican citizens who were also lawful permanent residents of Texas. Rovirosa owned, controlled or was associated with various energy companies (Mexico Energy Companies) active in the oil and gas industry. Avila worked for the benefit of Rovirosa and the Mexico Energy Companies and assisted in the payment of bribes to PEMEX officials, according to the criminal indictment filed against the Defendants (Indictment) in the Southern District of Texas (S.D. Tex.).

Criminal Charges, Trial and Conviction

On August 6, 2025, a grand jury in the S.D. Tex. indicted Rovirosa and Avila on three counts of aiding and abetting violations of the FCPA and one conspiracy count. With respect to the FCPA charges, the Indictment alleges that the Defendants used interstate commerce to “willfully and corruptly” aid and abet a “domestic concern” (which includes a U.S. resident) to facilitate payments to foreign officials in order to influence their official decisions, to induce them to violate their duties, to secure an improper advantage, and to induce them to use their influence to help Rovirosa and Avila to obtain and retain business for, and direct business to, Rovirosa and the Mexico Energy Companies.

Due to Avila’s status as a fugitive, the S.D. Tex. severed the trials of the Defendants. Rovirosa’s criminal trial began on December 1, 2025. Four days later, on December 5, 2025, the jury convicted Rovirosa on the conspiracy count and two of the three aiding and abetting counts. Rovirosa has since challenged the verdict by filing a motion to dismiss the case and a motion for judgment of acquittal.

Bags, Watches and Even a Treadmill to PEMEX Officials

From approximately June 2019 to October 2021, Rovirosa was complicit in three bribery schemes related to Petróleos Mexicanos (PEMEX), the state-owned Mexican oil company, and its wholly owned exploration and production subsidiary, PEMEX Exploración y Producción (PEP), DOJ prosecutors alleged in the Indictment. The government’s case was supported by numerous WhatsApp messages, originally written in Spanish and translated into English in the Indictment.

One scheme related to an audit of a pipeline maintenance contracts with three of the Mexico Energy Companies. In exchange for a favorable resolution of the audit, the defendants allegedly arranged to pay PEMEX officials approximately $130,000 in cash, a Louis Vuitton bag and a Hublot watch. In a WhatsApp message written in Spanish to the audit manager at PEMEX, Avila referred to the Hublot as a “commission haha.” The resolution of the audit reduced the payment owed by the Mexico Energy Companies to PEMEX/PEP by more than 75 percent. The audit manager messaged Avila that it was “mission accomplished!!!” and that he (the audit manager) “deserved a gift . . . what we agreed on.” After receiving the Louis Vuitton handbag, he thanked Avila, who responded, “you’re welcome bud; you’ve earned it.”

In another scheme, the Defendants arranged for the payment of cash bribes of at least $11,000 to the audit manager and a PEMEX procurement coordinator for the approval of a road and platforms contract that had a value of approximately $1.6 million. In furtherance of the deal, the audit manager messaged Avila that he had “ordered” that the audit be “resolved as soon as possible” so that PEMEX/PEP would “no longer have the audit as a pretext for not giving [the Mexico Energy Companies] work.” In reference to the bribe payment, Avila also messaged the audit manager, asking him to “tell me how we should divvy it up,” and adding that he would give the procurement coordinator “ten for sure.” After the award of the contract, the audit manager sent Avila a WhatsApp message stating, “mission accomplished, Bro!!!”

In a third deal, the Defendants promised to provide bribes, including a treadmill, to PEMEX officials to secure the awarding of a mechanical integrity contract with a value of approximately $2.5 million to two of the Mexico Energy Companies. To facilitate the bribe, the audit manager messaged Avila, “how can I help you with this?” Avila responded, “have them give it to him.” During the bidding process, the manager said that he was “looking into it” but needed to “start seeing some love already.” Although the bids of the Mexico Energy Companies were initially determined to be noncompliant, that decision was reversed after a meeting among the audit manager, the procurement manager and other employees. The audit manager then sent Avila another WhatsApp message again stating, “mission accomplished Bro!!!” adding that he had been a “damn bullfighter” who had “entered the ring!!” He told Avila that they would soon be “opening up the champagne.” A couple of weeks later, after the contract was awarded to the Mexico Energy Companies, the audit manager informed Avila, who messaged in response that he would let Rovirosa know about the award so he could “schedule the support.”

The total value of the consideration transferred to the PEMEX officials in connection with the alleged corrupt transactions was approximately $150,000.

A “Typical” Scheme

“This is a very typical FCPA case,” Martin Weinstein, a partner at Cadwalader and a former federal prosecutor, told the Anti-Corruption report. “You have an individual who is alleged to have bribed officials of a state-owned enterprise, and the amount of money is fairly significant,” he observed. The DOJ is looking to prosecute individuals and Rovirosa was available as a defendant, so they not only had someone whom they could indict, but also someone whom they could try, he said.

The Rovirosa prosecution can be seen as a “typical” bribery case in many respects, agreed Ricardo Cacho, a partner at Von Wobeser y Sierra, based in Mexico City. It involves an important Latin American company that awards procurement contracts, and someone participating in bribes to get those contracts, he explained. The timing of the transactions, how the audit was resolved and the contract was approved, and the participation of an intermediary, all documented in a neat timeline of messages, is “very typical in corruption cases,” he said, almost “as if it were a movie,” he reflected.

Beyond the Dollar Signs

While the total sum of approximately $150,000 in bribes is not staggering, it was more than sufficient to warrant the DOJ’s attention.

“The amount of the bribe is not really the point,” Weinstein explained. “Whether it is $1,000 or $5,000 or $100,000, it is a crime,” he said. In any event, $150,000 is “real money,” he emphasized, and something “well within” the range he would have considered as a prosecutor.

Although $150,000 might seem small compared to the size of the contracts that were awarded, it is possible that the alleged amount constituted only a portion of the total bribes, Cacho speculated. Also, the figure appears proportionate considering that the public officials who were bribed in this case were mid-level officials, he assessed.

Aiding and Abetting Are Easier to Prove

The prosecution relied on conspiracy and aiding and abetting theories as opposed to alleging direct violations of the FCPA. That was a “tactical” decision that presented a stronger case to the judge and jury, Hugo Lopez-Coll, a partner based in Mexico at Sainz Abogados, suggested.

If one tries to prosecute the orchestrator of a criminal scheme there is a “lot to prove” compared with someone who merely supported it, Cacho noted.

Targeting the Individuals and Not the Companies

One priority highlighted in the Blanche Memo is for prosecutors to “focus on cases in which individuals have engaged in criminal misconduct and not attribute nonspecific malfeasance to corporate structures.” The Rovirosa case is a notable example of that focus in action.

The DOJ’s past efforts were more centered on sanctioning corporations, particularly non-U.S. corporations, and extracting large fines from them, Lopez-Coll explained. The prosecution of the Rovirosa case aligns with the DOJ’s announced intention to pursue not only corporations, but also to increase its focus on individuals, he said.

The Rovirosa prosecution could also signal a greater DOJ focus on companies that were previously thought to be under its radar, Lopez-Coll continued. In addition to marquee companies trading on New York stock exchanges and global companies, the DOJ might also be turning its focus to smaller, regional companies and their owners, he suggested.

The targeting of the Defendants as opposed to the Mexico Energy Companies is “in line with” the new DOJ guidelines, Cacho concluded. It is sensible to go after a company the size of Walmart, he said. However, prosecutions of small companies are not as impactful and, in such cases, it might make more sense to prosecute the individual and impose a jail sentence instead of imposing a fine that the company might not even be able to pay, he concluded.

Often in FCPA prosecutions, companies write a check and no one goes to jail, Weinstein said. Congress and others have criticized this as a “loophole by which companies get to buy their way out of criminal conduct,” he explained. “If you really want to have deterrence, you have to send people to jail,” he advised. If the companies were solely owned by Rovirosa, there is not much to be gained by indicting the companies, as well, he noted.

See “Zaglin Conviction Offers Insights on Individual Prosecutions in Trump 2.0” (Oct. 22, 2025); “The Blanche Memo’s Take on Corporate Responsibility: Individuals Versus Corporations” (Sep. 10, 2025).

An Unlikely Trial

That Rovirosa’s case proceeded to trial is a noteworthy aspect of the case.

It is “very uncommon” for FCPA cases to go to trial because of the potential liability and reputational damage, Lopez-Coll said. In many cases companies agree to cooperate in investigations to have a better position before the DOJ, he explained.

However, given the DOJ’s renewed focus on prosecuting individuals, we might see more cases going to trial. An individual does not have the same incentives to avoid trial as a company, Lopez-Coll emphasized, nor do they have the resources to conduct an internal investigation or provide additional information to the DOJ. Instead, individuals are more likely to challenge the evidence presented by the DOJ, he noted.

Contesting the Verdict

Subsequent to the jury verdict, Rovirosa filed a motion for judgment of acquittal and a motion to dismiss the case.

An Uphill Battle

As an initial matter, overturning jury verdicts is usually an “uphill battle,” and judges are hesitant to do so, Weinstein said. Arguments about the sufficiency of the evidence “almost always lose on appeal,” he explained. Generally, the better argument is that information was introduced without a sound evidentiary basis, he said. Such arguments would go to whether the text messages were authenticated and the interview transcript was properly introduced into evidence, he stated.

Allegedly Improperly Admitted Evidence

In the motion for judgment of acquittal, Rovirosa argued that the government failed to present sufficient evidence to establish the conspiracy and aiding and abetting claims, and that the government presented evidence to the jury that was either not properly admitted into evidence or should not have been admitted. With respect to the latter, Rovirosa argued that the jury was improperly allowed to hear a recording of an interview between him and ICE agents that was not admitted into evidence and contained hearsay statements by Avila. He also argued that the jury was provided with text messages with alleged co‑conspirators that were not properly admitted into evidence.

No Witnesses to the Crime

Whether the evidence was admitted properly or not, Rovirosa additionally protested his conviction partly on the ground that the prosecution did not present the testimony of any witnesses with direct knowledge of the commission of the alleged crimes. Instead, the DOJ presented as witnesses Homeland Security Investigations special agents and a computer forensic analyst who testified about the investigation, a lawyer who testified about Mexican law, and a former PEMEX employee convicted of taking bribes in an unrelated case who testified about PEMEX’s procurement processes.

The DOJ, for its part, alleged that Rovirosa and his co-conspirators “largely orchestrated” their crimes by utilizing WhatsApp messages and relied exclusively on documentary evidence to prove its case.

“Can someone be convicted based entirely on documentary evidence? No question,” Weinstein explained. “If it is properly admitted, there is nothing that says the prosecution must have a live person with firsthand knowledge,” he said. Most of the time there will not be a person with firsthand knowledge of bribes being paid, except the defendant who is under no obligation to testify,” he noted.

Prosecuting cases based on documentary evidence is typically the way that FCPA cases are litigated in the U.S., Lopez-Coll said. Many are based on forensic analysis of emails and text messages, he continued, because it is difficult for the government to secure foreign witnesses to testify. If foreign officials with no connection to the U.S. receive a subpoena, their lawyers’ initial reaction will be “don’t answer, don’t be concerned,” Lopez-Coll explained. Here, what was important was the prosecution’s ability to tie the WhatsApp messages to banking transactions. It told a story that had a lot of weight and resonated with the jury, he said.

Not having a direct witness to a bribe is not unusual, Cacho confirmed. Many cases are prosecuted on almost completely documentary evidence, he said, especially in white-collar criminal cases. In this case, the documentary evidence was “pretty straightforward” and built a very coherent storyline, he added.

In the U.S., there are more incentives for witnesses who participate in bribery schemes to cooperate than in Mexico, Cacho explained. If a Mexican witness strikes a deal with U.S. prosecutors and testifies, they could expose themselves to prosecution in Mexico, and therefore are more likely to “deny everything,” he said.

Congruence With the Blanche Memo

The grand jury returned the Rovirosa indictment fewer than two months after the DOJ issued the Blanche Memo. The timing raises the question of whether the government brought the Rovirosa case to highlight its agenda or whether the case was simply ripe for prosecution.

The Rovirosa prosecution was a “great case” for the DOJ to signal its priorities under the Trump administration, Lopez-Coll said, but it was also “low hanging fruit” because the DOJ had the benefit of WhatsApp conversations that documented a detailed timeline of the bribery and the motivations for the payments.

The answer is not so clear, according to Weinstein. The DOJ may have chosen the case to signal priorities, but, more likely, it chose to proceed with the case because it was ripe for prosecution, he posited. “My guess is that they brought the case because they thought that this was a bad guy who did bad things.”

Protecting Opportunities for U.S. Companies

The Blanche Memo sets forth a series of “factors” that prosecutors are to consider when deciding to bring enforcement actions, one of which is safeguarding “fair” business opportunities for U.S. companies.

The FCPA helps U.S. companies to compete “successfully and legally” around the world, Weinstein asserted. U.S. companies are the “most compliant” with anti-corruption laws of any group of companies in the world, he posited. “If one looks at the history of the FCPA in the last 10 or 15 years, the majority of really big settlements were against foreign companies that were issuers in the U.S.,” he noted.

One of the motivations for the prosecution is that the alleged bribery potentially pushed out a good U.S. company that “could deliver great service and do things right,” according to Cacho.

The case appears to be an example of protecting fair competition in the energy industry, Lopez-Coll concluded. The case makes the point that tackling corruption in Mexico is necessary to protect U.S. companies competing in a strategic industry, he said.

See “The FCPA Lives: Protecting American Interests” (Aug. 13, 2025).

Potential Difficulties in Bringing Individual FCPA Actions

The DOJ’s stated focus on prosecuting individuals over corporate structures may face some unexpected difficulties.

Traditionally, many FCPA violations are brought to the DOJ’s attention either by whistleblowers who are incentivized to report violations or through self-disclosures, Lopez-Coll explained. However, an increased focus on prosecuting individuals instead of companies might reduce whistleblower incentives and, without this tool, it will be more difficult for the DOJ to bring cases, he surmised.

In addition, the focus on individual prosecutions might reduce the incentives for company self-disclosures, especially if high-level company officers could be targeted, Lopez-Coll suggested. DOJ investigations have traditionally focused more on the failure of a company’s controls for preventing corruption and the lower-level employees who paid the bribes, he explained. Companies were expected to cooperate with the DOJ, pay a fine without admitting criminal liability and perhaps be subjected to a monitorship, he continued. Now, however, a CEO, CFO or other officer who is potentially subject to criminal liability may no longer have incentives that are aligned with those of the DOJ, making them more reluctant to come forward due to the risk of prosecution, he concluded.

See “Do the 2025 Changes to the DOJ’s CEP and Whistleblowing Programs Encourage Companies to Self-Report?” (Jul. 16, 2025).

Compliance Considerations

The Rovirosa case offers several lessons for companies seeking to avoid FCPA liability.

Establish and Maintain a Robust Compliance Program

Most basically, a company should establish and maintain a strong compliance program, Weinstein said. The program should be tailored to the company’s actual risks and updated regularly to respond to changes in business and enforcement priorities, he noted.

A compliance program should ideally “introduce meaningful friction in high-risk areas that makes it harder to engage ‘fixers’ and compensate third parties without documented evidence of legitimate services and deliverables,” Lopez-Coll said. This is necessary to ensure that the program is “not merely well-designed on paper, but functions effectively in practice,” he explained.

Review Third-Party Risk

The Rovirosa decision is a reminder that companies should be mindful of the risks posed not only by employees but also third parties.

“Many issues arise from intermediaries, so strong diligence is key,” according to Weinstein. These include agents and distributors, he said. “Targeted training is also important, especially for employees and third parties in high-risk roles and regions,” he added.

“Companies should revisit their third-party engagement protocols, with particular emphasis on intermediaries acting on the company’s behalf before government clients or government agencies,” Lopez-Coll advised. They should require “documented and risk-based business justifications explaining why a particular intermediary is necessary,” he recommended. “Engagements based primarily on an intermediary’s government relationships or connections should be treated as heightened risk indicators and, absent legitimate and documented necessity, should not be approved,” he cautioned.

Due diligence should extend beyond basic background checks and include “red flag” monitoring for intermediaries. Organizations should also ensure that all third-party payments are conditioned on “legitimate deliverables rather than vague ‘success fees,’” Cacho said.

See “Conducting Effective Third-Party Due Diligence in Latin America” (Mar. 1, 2023).

Maintain and Enforce Messaging Controls

The bribery scheme in the Rovirosa case was committed largely through WhatsApp messages, underscoring the need for companies to establish and maintain strict controls for company communications.

“Companies must enforce clear policies regarding the use of personal messaging apps,” Cacho said. They should also utilize available tools to archive business communications, which will prevent “‘off the books’ negotiations” from being hidden from a company’s compliance officers, he advised.

See “Managing Off-Channel Communications in Internal Investigations” (Jan. 28, 2026).

Monitor Payments and Expenses Scrupulously

Companies should also maintain “tight financial controls” for vendor payments and expense approvals, including testing procedures to validate transactions, Lopez-Coll advised. Vendor payments should require contemporaneous documentation of the services provided, he stated. Further, these payments should be independently reviewed by employees who are not responsible for managing the relationship, and “companies should prohibit the purchase, payment, or reimbursement of luxury goods and similar high-risk items, particularly in connection with government-facing activities,” he emphasized.

“Companies should move toward real-time, data-driven expense monitoring,” Cacho agreed. Companies should have value limits and a pre-approval process for any gifts, travel or entertainment provided to government officials, he added. The company’s compliance team should also examine expenses for any concerning spikes in gift, travel and entertainment expenses immediately before or after the award of any contract or resolution of any audit, he advised.

See “Continuous Spend Monitoring for End-to-End Third-Party Risk Management” (Dec. 11, 2019).

The use of AI systems is transforming the way that compliance investigations are conducted, but it also raises issues of ethics, transparency and accountability in algorithmic decision-making. AI systems for compliance must be trustworthy and explainable, while balancing the drive for innovation with regulatory requirements. During a panel discussion at the New York City Bar’s International White Collar Crime Symposium 2025, international experts discussed these issues and offered guidance on how compliance professionals can best use AI tools, including agentic AI tools, and provided best practices for interacting with regulators.

The panel was led by Matthias Gstoehl, a partner at Schellenberg Wittmer in Switzerland, and included Avik Biswas, a partner at Khaitan in India; Stephane Eljarrat, a partner at Osler in Canada; Jitka Logesova, a partner at Wolf Theis in Austria; Patrick Manion, a manager at Harvey, an AI legal services startup; and Silvia Martina, a partner at Cagnola in Italy. This article summarizes their insights.

See “Benchmarking AI Uptake by Compliance Functions” (Dec. 3, 2025).

The Benefits of Using AI in Investigations

The main benefit of AI in an investigation is speed, Manion said. AI programs such as Harvey are useful at the preliminary stages of international investigations to translate foreign language documents in order to give the investigator an early picture of the investigative landscape. In addition to being much faster than hiring a translator, it is much cheaper and does not involve dealing with someone who may be in a different time zone.

In document reviews, a critical and cumbersome element of internal investigations, AI products are not just faster than running a keyword search, but also easier to use because they are natural language-based, Manion observed. AI tools do not require the user to input “algebraic equations” and “50‑word strings with terms and connectors and plus and minus signs,” he added.

AI tools work best in popular languages like English, German, Spanish and Chinese, but are less reliable – because they have had less training – in languages used by fewer people, such as Slovak, Croatian and Hungarian, Logesova warned.

AI was very useful in a case involving the review of about 12,000 different agreements, Eljarrat reported. It was used to create chronologies, timelines, summaries of witness interviews and a list of the cast of characters.

Another very useful AI function is sentiment analysis, in which the program reviews emails and seeks out unusual language patterns that indicate stress or a change in relations between people, according to Eljarrat.

Eljarrat’s firm has also made use of what is known as relationship investigations. For example, in a case his firm worked on, two individuals had stated that they had never met, she recounted. The AI program went through public web sources to find a photograph that showed, in the background, the two people sitting together at a table, thus disproving their claim of never having met. “This would have been impossible, obviously, to do in any other way,” he said, while noting that this type of relationship investigations could raise privacy concerns.

AI has proven itself useful in privilege reviews, as well, Eljarrat continued. In one case, a vendor stated that there were no privileged emails in a folder, but when the AI program checked the folder, it found that 35% of its emails were privileged.

Being able to show stakeholders that a company is using AI ethically builds trust and ensures accountability about decisions, Martina asserted. This requires a company having a supervisory body to oversee the transparency and traceability of automated decision-making systems, and to ensure algorithm security and the proper level of human intervention, she said.

See “AI Governance: Striking the Balance Between Innovation, Ethics and Accountability” (Jun. 18, 2025).

Reasons for Caution

While there are clear benefits to integrating AI into internal investigations, there are significant concerns, as well.

Anyone using AI must be aware of its limitations, which include hallucinations and sycophancy, Gstoehl warned.

AI users should be aware of the issue of biased and discriminatory results, Martina added.

Users of AI tools should also be aware that this technology is evolving and changing faster than the laws that regulate it, Eljarrat noted.

Additionally, there are regulatory, safety and training concerns of which investigators should remain aware.

European Regulations Around AI Use at Law Firms

There are three types of European AI regulations relevant to law firms and inside counsel using AI tools in investigations, Logesova explained.

Many European national bar associations – including those of the Czech Republic, Germany and Austria – have developed rules for lawyers using AI, Logesova said. The rules generally require lawyers to use AI tools responsibly; a main focus is ensuring that lawyers retain ultimate responsibility and control over AI output.

European lawyers also have data protection and privacy obligations under the General Data Protection Regulation (GDPR), a E.U. law that prescribes how companies deal with the personal data of European residents, Logesova said.

Article 4 of Regulation 2024/1689 of the E.U. Artificial Intelligence Act stipulates that law firms are only allowed to use AI if the law firm’s employees and partners are competent in its use. It is important that law firms be able to show they have provided training about AI risks, have developed binding and clear guidelines for AI use, define responsibilities for AI and not only have AI rules but are actually enforcing them, Logesova explained.

Companies must document everything about the use of an AI tool in an investigation and maintain audit trails, Eljarrat advised. It is also important to get the assistance of computer specialists to secure all metadata and to make sure that the AI tool is properly retaining data, he suggested. Then, the company should be prepared to be fully transparent with the regulator, explaining what AI tools were used on what data.

Difficulties in negotiating with regulators may arise if a compliance team is unable to explain exactly what an AI tool was “thinking” when it produced its outputs, Logesova said, noting that her firm uses the AI-powered Relativity platform.

See our two-part series: “AI Meets GDPR”: EDPB Weighs In on AI Models (Feb. 26, 2025), and Mitigating Risks and Scaling Compliance in the Development and Deployment of AI Models (Mar. 12, 2025).

Safety Concerns

AI can not yet be safely used for compliance tasks that involve high stakes judgments, gray areas and ambiguity, all of which can appear in an investigation, Martina said. However, that leaves plenty of room for companies to safely employ AI for more rote compliance functions.

AI can be safe and useful for compliance work in situations where there is a large volume of data, clear rules and repeatable patterns, Martina said. Thus, compliance tasks – like document reviews during investigations, monitoring emails for key words and transaction monitoring – are all jobs well-suited to AI. Third-party due diligence can also be safely assisted by AI tools that can help with background checks and identifying politically exposed persons.

Other tasks, such as risk assessment and mapping, can employ AI but require humans to remain in the loop, Martina said. AI can help to monitor behavior patterns to determine if rules are being ignored or followed and can also be useful in updating compliance polices after regulatory changes, she added.

See “How Under Armour and People Inc. Took AI Governance From Crawl to Walk to Run” (Oct. 8, 2025).

Training and Effective Use

Investigative teams using AI must establish a baseline level of AI literacy before employing AI tools in an investigation. Investigators need to be properly trained on prompting the AI and understanding the strengths and weaknesses of AI models, Manion said. Those who use AI at work should also use it in their personal lives, to increase their general level of AI fluency, he suggested.

In-house compliance investigators play a crucial role in working on sensitive matters and in assessing the outcome of an investigation, according to Martina. Providing advice to the company’s board on whether to file a criminal complaint as a result of an investigation or whether improvements to the company’s compliance are necessary should be left solely to humans, she stated.

Clients at Eljarrat’s firm are demanding that its lawyers use AI in investigations to diminish costs, he said. The main people using AI at a law firm should be associates and IT discovery staff, who should be trained as to the risks and limitations of the technology. Partners should not be the primary users of AI tools “unless [they] want to get fired by [their] client,” he quipped, noting that partners should work on tasks involving judgment, strategic thinking and decision-making. Associates should handle AI work as a way to gain experience, he suggested. All AI users at a firm should undergo continuous training with regular meetings to update them of any changes in the investigation or how the tool is being used. Compliance professionals who do not meet regularly with their team to update and refine AI prompts could “end up with a potential disaster.”

However, the most junior of associates at a firm should not be allowed to use AI tools, either, Logesova argued. At her firm, only senior associates and above can use them, in order to reduce the risk of junior associates using the tools irresponsibly. It is also important for the data used by AI tools to be properly prepared before analysis by an AI tool, she said, adding that this is an area in which her firm has determined that it needs to speed up its processes.

Agentic AI and Compliance

Agentic AI is a powerful tool for compliance professionals, with risks that can be minimized by understanding the nature of the technology and by following certain best practices.

Comparing Agentic AI to Generative AI

Generative AI produces content by predicting patterns based on training data, while agentic AI consists of systems that orchestrate various other data systems to solve problems with reduced human input, Manion explained. Generative AI is typically used for tasks such as summarizing documents and drafting memos or emails. Agentic AI, however, is primarily used for tasks such as proactive research or scanning for new regulatory changes.

Use of generative AI typically involves providing it with a prompt for a single task, such as drafting an email, while agentic AI generally involves a prompt consisting of a problem statement, rules and a goal, which the agentic AI will split into different tasks and sub-tasks as it seeks a solution, Biswas explained.

To successfully use generative AI, the most important skill is writing prompts, but to most effectively use agentic AI, one should interact with it in much the same way as one would assign a task to a colleague, Manion advised.

Agentic and generative AI are both subject to the “garbage in, garbage out” rule, which means that unless these systems are provided with accurate, reliable information, they will produce inaccurate and unreliable results, Biswas noted.

Agentic AI Best Practices

Agentic AI has limitations with regards to audit trails, particularly in terms of being able to trace the information used and the steps taken to produce the output, Eljarrat said. Major law firms around the world are still trying to figure how to implement and test these tools, he noted.

To stay on the safe side, compliance teams should not use agentic AI in cases in which litigation is anticipated, as its output may not be admissible as evidence in court, Eljarrat advised. Agentic AI should also be avoided in complex investigations involving data from multiple jurisdictions because of privacy and other regulatory risk, particularly in the E.U. There are also risks associated with using agentic AI in privilege reviews, a task requiring judgment and nuance, he added.

When using agentic AI, it is important to keep humans in the loop for at least two reasons, Biswas suggested. First, the agentic AI may miss important nuances. For example, an agentic AI might flag an employee for apparent excess spending, without realizing that this employee is a traveling salesperson whose job requires much higher expenses than other jobs at the company. Second, human involvement makes it more difficult for a regulator or a litigator to claim that the results of the agentic AI’s work was just a mindless “rubber stamp,” he noted.

See “Integrating AI Into the Five Stages of an Investigation” (Oct. 8, 2025).

Live, in-person trainings are considered the gold standard for employee learning, but most companies now use some form of prerecorded and asynchronous education programs. These “click-through trainings” can be delivered to a broad audience that only has to click through screens and multiple-choice tests to complete the training. Click-through training has significant advantages over traditional training methods and can be a powerful tool for keeping a large and far-flung workforce informed, but getting the most out of it requires thoughtfulness and planning.

This quick start guide sets forth the basics of how to fit bite-sized learning experiences into a broader training program.

For a more detailed discussion, see our three-part series “Rethinking Click-Through Training”: The Pluses and Minuses (Feb. 26, 2025), Maximize Effectiveness With Customization (Apr. 9, 2025), and Integration Into a Comprehensive Training Program (May 7, 2025).

Understand the Advantages and Disadvantages

When deciding how to employ click-through training, it is important for a company to be clear-eyed about the benefits and drawbacks of this type of learning.

Pluses

• Efficiency: As self-guided, computer-based learning, click-through training allows organizations to distribute training to large groups without a massive resource burden.
• Convenience: Click-through training allows employees to complete sessions at their convenience within a given time frame, making it more flexible than scheduled in-person training. Trainers also have the advantage of recording a lesson whenever it works best for them, without having to fly to far-off locations to train teams on the ground.
• Consistency: Click-through training also helps ensure that a consistent compliance message is delivered to employees in multiple regions.
• Familiarity: Most medium and large companies use click-through training, making it a familiar option to both employees and regulators.
• Immediate Feedback: Click-through trainings offer an advantage over in-person lectures by allowing immediate knowledge checks, ensuring that employees actually engage with the material in some measure rather than passively listening.

Minuses

• Lack of Engagement: The term “click-through training” itself suggests training is a check-the-box exercise, potentially giving people license to just click without engaging with and thinking critically about the content.
• Too Many Trainings: Click-through trainings are frequently deployed as one of a number of mandated trainings that must be achieved within a designated time frame, leading to training overload.
• Repetitiveness: Annual training requirements on the same topic do not necessarily mean that a different or improved training course will be offered subsequently. An employee may end up taking the exact same online course year after year, which could potentially lead to training fatigue and disengagement.
• Completion Over Comprehension: Many companies focus on ensuring that employees take the training without inquiring into whether it has had any real impact on behaviors or understanding.

See “How Nadège Rochel of Hollister Inc. Uses Monopoly, a Deli Counter and Emotional Intelligence to Promote Compliance” (Aug. 7, 2019).

Develop Good Content

For compliance training to be effective, a company must first determine its training objectives and customize trainings to the company and different roles within the company, where appropriate.

Meet Multiple Training Objectives

• Building Relationships: Because training is a key way the compliance team engages employees, the interaction should be welcoming and introduce the team while highlighting the resources it offers.
• Behavior Change: Training is meant to educate employees about company policies but also needs to deter bad behavior and encourage employees to speak up if they see something wrong.
• Nuanced Decision-Making: In the real world, bribery solicitations may take subtle forms such as vague success fees, ambiguous tolls or indirect benefits like internship offers for a customer’s child. Thus, training also needs to help employees make nuanced decisions in these gray areas.

Customize to the Company

• Branding: The most basic options for customization focus on adding company branding – logos and color schemes – to off-the-shelf training materials. Incorporating branding into click-through training can reinforce familiarity and make compliance feel like an integral part of the corporate culture rather than an external mandate.
• The Right Terminology: Click‑through training should use terminology employees are familiar with. Make sure that the policies and procedures being referenced are the titles the organization actually uses.
• Familiar People: Having current leaders and employees appear in content, whether in videos or other images, can make a click-through training resonate with workers far more than actors spouting compliance platitudes from staged settings.
• Real Events: Training is most effective when it includes real, company-specific examples such as success stories of employees who identified fraud or internal cases where noncompliance led to consequences. If an organization is not comfortable talking about its own prior problems, it might instead address challenges experienced in its industry.

Customize to the Employees

• Differing Roles: Employees at different levels and in different roles have different learning needs, and training should be designed with that in mind.
• Cultural Diversity: Cultural differences can be a hazard for corporations trying to provide compliance training on a broad scale, so trainings should be adjusted to this diversity wherever possible.
• Provide Variety: A strong compliance program is adaptable and should ensure that employees do not receive the same training year after year. Instead, each training session should include fresh content, updated scenarios and relevant case studies.

See “How Ericsson Made Compliance Training Must-See TV” (Mar. 12, 2025).

Measure Effectiveness

Regulators have made clear that they care less about the form and content of compliance programs in general – and trainings in particular – and more about effectiveness, but measuring efficacy can be tricky.

• Test Understanding: The most obvious way to assess effectiveness is to test employees after a training to determine what they learned. These tests can highlight places where the training might need improvement.
• Assess Growth and Retention: Effectiveness can likewise be assessed by providing a pre-training test as well as a post-training test to demonstrate what people used to know and what they know as a result of the training. A company can also test six weeks or six months later to see if learners retained the knowledge.
• Measure Behaviors: Other key metrics for evaluating training effectiveness focus on post-training behavior changes, including whether the training results in fewer policy exceptions or rule deviations.

See “Using the Kirkpatrick Model to Measure Healthcare Compliance Training Effectiveness” (Nov. 19, 2025).

Integrate Into a Broader Training Program

To deliver and reinforce long-lasting behavioral change, click-through trainings need to be embedded within a broader compliance training framework that mixes asynchronous online learning with synchronous online training, in-person sessions and leadership engagement.

• Synchronous Online Training: Because click-through training is not the best place for employees to work through trickier situations, live trainings, where employees can interact with their colleagues and knowledgeable leaders, are more impactful when there are more nuanced issues to be addressed.
• Face-to-Face Interactions: In-person, face-to-face trainings have benefits beyond the training itself. In a live setting, the instructor not only conveys information, but also learns from employees.
• Timing: Training should not just be an annual or end-of-year event. One benefit of online training is that it can be offered at the very moment it is most relevant. If, for example, an employee has to fill out a particular form, a quick pop-up click-through training on how to fill out the form properly can be very effective.
• Test-Out Options: To be respectful of employees’ time and intelligence, companies can offer test-out options so a learner who has retained a training’s content does not have to retake it as a “refresher” every year.
• Incorporating AI: One of the primary drawbacks of click-through training is that a learner cannot ask questions or for clarification, but an AI chatbot can fill that gap.

See “High- and Low-Tech Innovations to Overcome Compliance Training’s Drawbacks in the Financial Industry” (Jan. 10, 2018).

Choose a Vendor

Choosing the right vendor is one of the most important decisions when introducing click‑through training.

• Consider Internal Limitations: For many compliance teams, the learning management system through which a compliance training is deployed may be predetermined, which, in turn, may limit vendor choice and restrain flexibility in training format.
• Determine Appropriate Size: For large companies in search of training for large workforces, a large vendor may be able to offer much-needed consistency and infrastructure. Smaller vendors can provide more flexibility and cost-effective customization.
• Test for Freshness: It is important for vendors to periodically refresh content and visuals to maintain engagement. A training created by a vendor should not have the same look as it did five or 10 years ago.
• Negotiate Wisely: It is important to understand how each vendor’s pricing works and what is included in that price. Many vendors use a licensing model, charging for each employee that takes a training, but they may be open to different pricing structures. Also check whether content updates are included in the price.

See “NAVEX Global Ethics and Compliance Training Survey Offers Benchmarking Data on Compliance Training” (Aug. 16, 2017).

 

This Quick Look incorporates original reporting by Lori Tripoli.

Artificial intelligence (AI) – and generative AI (Gen AI) in particular – is being incorporated at an unprecedented rate into virtually all areas of the economy. Thus, it is not surprising that virtually all fund managers who participated in a study conducted by the Alternative Investment Management Association (AIMA) said they use Gen AI in their work – and most are increasing their use. Earlier this year, AIMA asked fund managers about their uptake of Gen AI, including their approaches to governance, policies and procedures, risks and limitations of Gen AI, training and hiring, Gen AI models, and use cases. It also asked institutional investors about their AI-related concerns and communications with fund managers. This article distills the key takeaways from AIMA’s study.

See “A Baker’s Dozen AI Governance Resolutions for 2026” (Jan. 14, 2026).

Demographics and Methodology

AIMA surveyed 150 fund managers, with approximately $788 billion in assets under management (AUM). AIMA’s survey report (Report) breaks down responses between managers with more than $1 billion in AUM (large managers) – which accounted for about 60% of respondents – and those with less (small managers). Respondents were based about equally across the U.S., U.K. and Asia-Pacific region.

Additionally, AIMA surveyed 18 large institutional investors, including U.S. state pension funds, endowments and family offices from around the globe. AIMA supplemented the surveys with multiple qualitative interviews and roundtable discussions with fund managers, investors and service providers.

The Report is organized around the following considerations for effective and responsible implementation of AI, with relevant takeaways for managers:

• implementing strong governance;
• adopting appropriate policies and procedures;
• understanding the risks and limitations of AI;
• training;
• selecting the appropriate AI model;
• identifying use cases;
• developing potential uses in the front office;
• understanding investors’ concerns and communications; and
• using agentic AI.

AI Governance

Broad Adoption of AI

Virtually all managers (95%) said they use Gen AI in their work, including 75% that have increased their use of Gen AI. Sixteen percent of managers use Gen AI only through internal tools, a cohort consisting primarily of large managers. Such internal tools help firms ring-fence sensitive data, notes the Report.

Large Managers Have More Controls Over AI

Large managers generally impose more restrictions on using external Gen AI tools than small managers. For example, 62% of managers overall permit external AI tools subject to policy parameters – including 75% of large managers and 41% of small managers. Conversely, one-quarter of managers impose no restrictions on using external AI tools – a group consisting primarily of small managers.

Nearly half of managers restrict the types of Gen AI tools employees may access, including 63% of large managers and 24% of small managers. Most managers (81%) restrict input of sensitive or confidential data, including 89% of large managers and 69% of small managers. More large managers than small managers also restrict using Gen AI for:

• tasks related to investment strategy (57% vs. 33%);
• generation of marketing content (41% vs. 10%); and
• compliance-related tasks (28% vs. 12%).

“The risks of ungoverned usage – such as reputational damage, regulatory breaches, or intellectual property loss – are not size-dependent,” cautioned AIMA. The “gold standard” for deploying Gen AI is a secure, in-house generative pre-trained transformer (GPT), an AI system that routes every AI request to an appropriate AI model, explains the Report. However, internal GPTs require significant resources for initial development costs and ongoing maintenance, operations and oversight.

Public vs. Private Models

Company-specific in-house AI models can be trained on firm-specific data and to tailor responses to the company’s particular needs. Many firms are deploying AI through a combination of secure third-party tools, purpose-built internal applications and open-access tools. Of the large managers that use Gen AI for work purposes, more than one-third (37%) use a public web-based version, while 28% use a company-specific in-house version and an additional 28% use both. In contrast, 73% of small managers use a public version, just 4% use an in-house version and 17% use both.

Takeaways

• Adopt clear AI use policies and protocols and conduct appropriate training.
• Encourage safe experimentation without stifling innovation.
• Tailor AI governance practices to address relevant risks – regardless of firm size.
• Understand common concerns, including data leakage and untrusted models.
• Consider combining both internal and external AI systems, as appropriate.

Policies and Procedures

Acceptable Use Policies Are Critical

The first line of defense for any firm that deploys AI is having a clear acceptable use policy, according to AIMA. Having and communicating such a policy can mitigate the risk of “shadow adoption” – employees’ using AI without the firm’s knowledge. In that regard, more than three-quarters of managers said they have an AI policy (53%) or are presently working on one (25%). On the other hand, 22% said they do not need a policy. A manager that permits staff to use open-access AI without a governing policy or appropriate restrictions faces a significant risk that sensitive information may end up on external systems where it could be used for training or included in answers to third-party queries.

Takeaways

• Adopt a clear, accessible Gen AI policy that sets expectations, defines acceptable use and prohibits risky behavior.
• Have visibility into how AI is used, including shadow adoption.
• Ensure appropriate training and oversight.

Risks and Limitations

Data Security and Hallucinations Are Top Risks

AIMA asked managers to list what they believe to be the three greatest risks associated with using Gen AI tools. Managers overwhelmingly cited data security and privacy (83%) and so-called “hallucinations” – false or fabricated outputs (64%). The next most commonly cited risks include:

• reliability (25%);
• training (17%);
• intellectual property infringement (15%);
• regulatory compliance (14%);
• lack of technical expertise (12%); and
• cybersecurity (12%).

Mitigation of Hallucination Risk

Although it may be impossible to eliminate hallucinations, there are ways to mitigate the risk they pose, notes the Report, including:

• connecting a model to trusted internal data sources;
• training a model with a firm’s own information;
• designing prompts to limit speculation and enforcing clear boundaries;
• requiring human review of outputs in high-risk and public-facing applications; and
• appropriate training and use guidelines.

“Human oversight of Gen AI is here to stay,” AIMA found. Managers know their firms remain accountable for issues caused by AI. Many said that full automation would not be suitable for front office operations or investor relations. Notably, too, quantitative strategy managers said AI lacks sufficient reliability for data analysis and executing trading models.

Takeaways

• Prioritize managing key risks.
• Take steps to reduce the risk of hallucinations.
• Ensure human oversight.
• Avoid full automation in “high-trust areas.”
• Ensure staff understand the limits of Gen AI.

Training and Hiring

Training

Most managers said they have either already conducted staff training on using Gen AI (26%) or are planning such training (44%). On the other hand, 24% do not plan to do any such training. Far more large managers (37%) than small managers (6%) have already conducted training. Moreover, 36% of small managers do not plan to conduct training, versus just 18% of large managers.

Of the managers planning training, a majority will use their own staff. Thirty-seven percent of large managers and 43% of small managers will use a third-party service provider.

Training does not have to be complicated, notes the Report. Basic training should cover:

• data leakage, hallucinations and other core risks;
• firm policies and use restrictions; and
• when to validate responses and escalate concerns.

More advanced training could include prompt engineering and custom GPT deployment.

Hiring and Consultants

Most managers said they are either not planning any AI-related hiring (53%) or are not likely to hire someone in an AI-dedicated role (18%). Just under one-fifth already have hired an AI specialist (15%) or are planning to hire one (4%). An additional 10% might hire one or more specialists. None of the managers in the study with less than $500 million in AUM reported having an in-house AI specialist. The main barriers to hiring AI specialists include high compensation and finding individuals with the right technical skills and/or financial acumen.

Roughly half of managers said they have not used external AI consultants to assist with Gen AI and do not plan to use them. One-third would consider using them. Nearly one-quarter of large managers have used consultants in a limited capacity, versus just 4% of small managers. Only 2% of managers – consisting entirely of large managers – have used consultants extensively.

Takeaways

• Conduct training when adopting AI – all firms, regardless of size.
• Start training with AI basics, especially safety.
• Consider hiring outside assistance.

Choice of Model

ChatGPT and Copilot Predominate

The two most common Gen AI tools mangers use are ChatGPT (81%) and Copilot (62%). Nearly one-fifth use Claude or Google’s Bart/Gemini. “All [large language models (LLMs)] will be capable of many tasks, but should not be viewed with [a] one-size-fits-all approach by those looking for best-in-class tools,” notes the Report. Different models have different strengths. Managers with custom GPT models enable users to connect to multiple LLMs. Private models have less risk of data exposure than open-access models.

Takeaways

• Understand models’ different strengths and choose the model best suited for a particular use case.
• Consider an LLM as a tool in a productivity suite rather than just a search engine.
• Consider models other than ChatGPT.
• Choose an internal vs. an open model based, in part, on the sensitivity of the data involved.

Common Use Cases

At least one-quarter of managers said they use Gen AI tools for:

• general research (65%);
• document analysis and summaries (62%);
• meeting summaries and minutes (40%);
• learning and professional development (38%);
• coding (31%);
• investment research and analytics (29%);
• investor relations and marketing (28%); and
• legal documentation (26%).

With respect to each of the use cases other than investment research and investor relations, a greater proportion of large managers than small managers reported using Gen AI for such purposes. At the other end of the spectrum, just 3% of managers said they use Gen AI for portfolio optimization.

AIMA also asked managers to identify the functions for which Gen AI tools have the most significant impact. The most common responses were:

• research (48%);
• marketing and investor relations (36%);
• IT (36%);
• legal and compliance (35%); and
• operations (26%).

Twelve percent cited accounting and finance and/or portfolio management. Just 5% cited HR.

Takeaways

• Encourage “bottom-up” experimentation with appropriate governance.
• Create cross-functional working groups to identify and share use cases.
• Be aware that some use cases might not deliver the expected impact.

Front Office Potential

Increasing Use for Investment Decision-Making

Managers are increasingly using Gen AI tools in the front office. Half said they use such tools for investment decision-making processes, up from 43% when AIMA conducted a similar study in 2023. In addition, 3% of managers overall, including 6% of small managers, use AI for portfolio optimization. Moreover, a majority of managers (58%) said they expect to increase use of Gen AI in their investment processes in the coming year, including 12% who expect to use it to “a great extent.”

The consensus among the interviewees was that AI can be used to augment – but not replace – investment professionals. Some recognized its potential for analyzing large volumes of unstructured data. In all cases, human oversight must be maintained.

Investor Concerns and Communications

Investors Increasingly Asking About AI

Most respondents said Gen AI is “a high priority discussion point” between fund managers and investors. In that regard, 29% of investors include specific Gen AI questions in their due diligence questionnaires, and a similar proportion plan to introduce them in the coming year. Of the investors that already ask questions about AI, most ask about:

• model oversight and explainability (82%);
• intellectual property risks (71%);
• data governance and privacy (65%); and/or
• regulatory or compliance concerns (65%).

Additionally, 60% said they would be more likely to invest with a manager that allocates a meaningful portion of its budget to AI research and implementation. However, just 18% ask about the impact of AI use on management fees. “Investors are divided on whether they would pay more for AI-enhanced performance,” notes the Report. Moreover, most investors expect AI-related costs to come out of managers’ existing research and development budgets. Consequently, managers should try to show how AI adds value – not just cost.

Concerns Over Long-Term Impact of AI

All investors agreed that Gen AI will give leading managers a competitive edge, but only half believe the edge will be sustainable. Most feel managers are not fully leveraging AI tools. On the other hand, “fund managers who talk up transformational AI strategies today will be expected to deliver measurable progress tomorrow,” said AIMA.

Forty-three percent of investors believe that, over the next three years, Gen AI will have a positive impact on the performance of the most sophisticated managers only. Just 13% believe it will positively affect most managers’ performance. Additionally, 38% believe its impact will depend on how the tool is implemented. Just 6% believe it will have little to no impact.

Takeaways

• Understand that investors demand “transparency [about AI] over hype.” Clear, realistic messaging is critical.
• Eschew “AI washing” and be transparent about experimentation stages and implementation timelines.
• Be able to deliver on AI-related promises and show what AI is actually accomplishing.

Agentic AI

The next stage of AI development involves “agentic AI,” which is defined as “systems capable of acting with autonomy, planning, and executing complex tasks rather than simply responding to prompts,” according to the Report. They operate “with a higher degree of independence, executing tasks and making decisions within predefined parameters.” Potential uses include:

• conducting risk and compliance monitoring, including monitoring positions and markets and conducting real-time surveillance;
• engaging in investment research and analysis, including scanning large datasets for signals and/or anomalies;
• optimizing and rebalancing portfolios within risk parameters; and
• automating routine workflows.

To prepare for agentic AI, fund managers should strengthen governance and accountability, train staff and ensure the integrity of the underlying data. Moreover, human oversight will remain essential.

See “Risk Assessment for Trump 2.0: Employing Data and Emerging Technologies” (Dec. 31, 2025).

Ogletree Deakins has welcomed Jane Norberg to the firm’s Washington, D.C., office, as a shareholder and co-chair of the whistleblower and compliance practice group.

Previously, Norberg spent nearly a decade at the SEC, where she served as deputy chief and then chief of the Office of the Whistleblower. In that role, she led the SEC’s whistleblower program established by the Dodd‑Frank Act of 2010. Most recently, she was a partner at Arnold & Porter.

Norberg’s practice focuses on proactively counseling organizations regarding whistleblower compliance obligations, including best practices related to emerging programs led by the SEC, Commodity Futures Trading Commission and DOJ, and leading internal investigations. She defends employers in government agency investigations, subpoenas and enforcement actions, and trains leadership teams on reporting and anti-retaliation programs. In addition, she has years of experience managing executive compensation and employee benefit matters.

For commentary from Norberg, see “NAVEX Study Finds Incident Reporting Steady, But Substantiation Rates Rising” (May 7, 2025).