FBI agent accused of stealing $1M in crypto and turning to AI for an escape plan
A former FBI supervisory special agent is at the center of a striking federal case after being accused of siphoning roughly $1 million in cryptocurrency from wallets tied to foreign adversaries, blending the funds with his own money, and consulting ChatGPT for advice on how to invest the windfall and move his family to Europe.
According to an affidavit filed on August 1, the ex-agent, identified as Yaroch, allegedly discovered private keys that unlocked access to several digital asset wallets described in court records as “adversarial cryptocurrency accounts.” Rather than following standard seizure or forfeiture procedures, he is accused of quietly transferring the assets to himself across about a dozen transactions starting in late 2024 or early 2025.
Prosecutors say the total amount taken was around $1 million, though the court documents do not specify which cryptocurrencies were involved or identify the precise wallets. What they do allege is that the transfers were executed for his personal gain, not as part of any authorized government operation.
Yaroch had served as a supervisory special agent in the FBI’s Counterintelligence and Espionage Division at headquarters, after previously working in the Boston field office. That role may prove pivotal in court: it could explain how he came into contact with the wallet keys and why he had privileged access to sensitive financial information tied to foreign adversaries. The affidavit, however, does not publicly disclose how those wallets were first obtained by the FBI or what specific investigation they stemmed from.
Federal investigators say that after moving the funds, Yaroch merged the disputed cryptocurrency with his existing holdings, effectively commingling the allegedly stolen assets with his personal finances. At some point during this period, he reportedly turned to ChatGPT for help in deciding what to do next.
According to the affidavit, his chats with the AI tool included questions about how to spend or invest $1 million and whether he should leave the United States for a European country. He shared details about his family, the kind of property he wanted-specifically a 2-5 hectare estate-and his interest in age-worthy red wine. In one cited response, ChatGPT recommended that he focus on relocation and lifestyle rather than chasing citizenship immediately, and singled out Portugal as the best fit for his circumstances.
“Given everything you’ve told me – [name of Yaroch’s child], your wife, the desire for a 2-5 hectare estate, interest in age-worthy red wine, and the goal of actually living there rather than just owning a property – I would not start by chasing citizenship,” the chatbot advised, according to the filing. The AI-generated suggestion reportedly highlighted Portugal as his top option.
Investigators later discovered that Yaroch had purchased airline tickets for a trip to Portugal, with a departure on September 3 and a booked return flight. Court records do not indicate that he followed any particular investment strategy suggested by the chatbot, nor do they claim that ChatGPT had any awareness the funds he was asking about were allegedly stolen. From the government’s perspective, the AI system was simply another tool he used while contemplating what to do with the money.
The affidavit also notes a series of admissions from Yaroch as the investigation progressed. He reportedly told a Department of Justice employee that he had made “some very poor decisions related to cryptocurrency wallets.” In a separate conversation with federal agents, he acknowledged that he had made a serious mistake, though the extent to which he confessed to specific actions is not fully detailed in the public documents.
The FBI terminated Yaroch’s employment on July 31. Soon after, federal authorities arrested him and charged him with interstate transportation of stolen goods, as well as receipt of stolen goods, securities, and money. Those counts remain allegations at this stage; he has not been convicted of any crime, and the case is still moving through the legal process.
What sets this case apart is not just the alleged theft itself, but the combination of insider access, cryptocurrency, and generative AI. Instead of an external hacker exploiting a vulnerability in software or hardware, the core of the accusation is that a trusted law-enforcement insider leveraged his official position to access wallet keys, then used consumer AI tools to game out how to enjoy the alleged proceeds and potentially start a new life abroad.
The timing of the incident coincides with heightened scrutiny around crypto custody and key management more broadly. Recent high-profile security events have forced the industry-and regulators-to confront how private keys are generated, stored, and accessed, and what happens when those controls fail.
One separate case drawing attention involved a flaw discovered in Coldcard’s seed generation process. Over a five-year span, that weakness has been linked to suspected attacks involving more than 1,800 bitcoin spread across over 5,200 potentially affected addresses. While research firms have cautioned that these numbers are based on on-chain analysis and do not definitively prove that a single attacker is responsible for all of the losses, the scale of the issue has underlined how a single technical or operational oversight can compromise thousands of users.
In another recent incident, the Ostium protocol reported that an attacker breached its off-chain infrastructure and manipulated BTC-USD price feeds. By distorting those price reports, the attacker allegedly drained 23.75 million USDC from one of the protocol’s liquidity vaults. Ostium emphasized that its smart contracts and governance multisignature wallets had not been directly compromised, underscoring that the weak point lay in the oracle and infrastructure layer rather than the on-chain contracts themselves.
Against that backdrop, the Yaroch case stands out as an example of an entirely different category of risk: insider abuse. Instead of a bug, misconfiguration, or external attack vector, the alleged problem here is a human being with legitimate credentials and elevated trust who chose to ignore protocol and policy. When one person is granted access to private keys or critical wallet information, all of the surrounding cybersecurity measures can become irrelevant if that individual decides to misuse their role.
For public institutions and private companies alike, this raises urgent questions about how access to digital assets is structured. Relying on a single person to hold or control sensitive keys-even someone highly vetted, such as an FBI agent-creates a concentration of risk. Many security experts advocate for multi-signature arrangements, strict role separation, and hardware-enforced policies to reduce the chance that one insider can move funds unilaterally.
The case also feeds into an emerging debate about how generative AI tools fit into the ethics and practice of sensitive professions. ChatGPT and similar systems are increasingly used by employees across finance, law, and government for drafting documents, brainstorming ideas, or researching complex topics. In this scenario, an AI chatbot apparently became part of an alleged scheme to plan the use of illicit funds and evaluate international relocation options.
From a legal standpoint, the use of AI does not absolve individuals of responsibility, but it could play a role in how investigators reconstruct intent. Search histories, prompts, and AI responses can provide insight into what a suspect was considering, when they considered it, and how premeditated their actions may have been. That kind of digital trail is already turning into a valuable evidentiary source in modern criminal cases.
At the same time, the affidavit’s explicit clarification that ChatGPT had no knowledge of the funds’ alleged origin highlights a broader reality: AI tools are not context-aware in the way humans might expect. They respond to the text they are given, not to the legal status of the money, data, or plans being described. This gap can create a false sense of neutrality that some users may try to exploit, though law enforcement can still retrieve logs and query histories when warranted.
For the cryptocurrency ecosystem, the allegations against a U.S. law-enforcement officer could become a reference point whenever regulators discuss the risks around digital asset custody. The narrative that crypto is “only” vulnerable to hackers is already outdated; insider threats, legal seizures gone wrong, mismanaged wallets, and flawed infrastructure have collectively caused billions in losses across the sector.
The case may prompt agencies that hold or manage digital assets-whether for investigations, sanctions enforcement, or asset forfeiture-to revisit their internal policies. They may look to enhance oversight, require multiple approvals for any movement of funds, increase logging and monitoring of all wallet-related activity, and limit the number of staff who can directly access private keys or seed phrases.
In the longer term, this incident could also influence how AI usage is governed inside law-enforcement and intelligence organizations. Institutions may choose to impose stricter guidelines on what employees can discuss with public AI systems, especially when it involves personal financial scenarios, foreign relocation, or operational details. Internal, audited AI tools with clearer logging and oversight might become more common in sensitive environments where external services are viewed as too risky.
As the legal process unfolds, the Yaroch case will likely be watched closely not only by criminal justice professionals but also by the crypto industry and AI policy circles. It sits at the crossroads of three powerful trends: the rise of digital assets, the growing reliance on generative AI, and the enduring challenge of controlling insider risk in high-trust organizations.
For now, the allegations paint a picture of an experienced counterintelligence agent who allegedly crossed ethical and legal lines, moved seven figures’ worth of digital assets, tested escape scenarios with the help of an AI chatbot, and booked a ticket to one of Europe’s most crypto- and expat-friendly destinations-only to be stopped by the same federal system he once served. Whether a jury ultimately agrees with that narrative will depend on evidence that has yet to be fully aired in court.