OpenAI Bans ChatGPT Accounts Behind Fake FBI Recovery Scam

OpenAI has banned a cluster of ChatGPT accounts that used its models to run a fraud-recovery scam built on fake law firms, impersonated attorneys, and a counterfeit version of the FBI's Internet Crime Complaint Center — an operation the company says very likely originated in Cambodia.

The company disclosed the takedown, dubbed Operation "False Witness," in its February 2026 report on disrupting malicious uses of AI. OpenAI's assessment of the activity aligns with recent public reporting on Chinese-led criminal scam operations operating out of Cambodia, a country that has become a global hub for industrial-scale online fraud.

The scheme targeted people who had already been defrauded — a second layer of predation that turns victims' desperation to recover lost money into fresh revenue for criminals.

The bait: six fake law firms

The scammers used ChatGPT to generate promotional content purporting to come from at least six fictitious law firms, according to OpenAI. Social media accounts and online ads then pushed these phony "scam recovery services" to potential targets.

The firms did not survive basic scrutiny. OpenAI's investigators found multiple open-source indicators of fraud: no evidence of state bar licensing, incongruous web domains, contact information listing street addresses that appear not to exist, and instructions to contact the "lawyers" only through messaging apps.

The fake persona problem ran deeper than the firm names. Scammers used ChatGPT to draft comments designed to emulate the tone and professionalism of attorneys who specialize in recovering financial losses for scam victims. Profile pictures for these invented lawyers appeared to be lifted from social media or generated with AI. In some cases, OpenAI found, the scammers recycled the same "attorney" identity across multiple supposed law firms.

One website in the network posed as the FBI's Internet Crime Complaint Center, better known as IC3. Clicking its "file a complaint" button routed visitors to an impersonation account on Telegram.

The hook: the International Criminal Court and a 15% fee

Once targets made contact on private messaging apps, the scammers deployed ChatGPT to manage the conversation. In messages drafted with the model, they falsely claimed to operate under the supervision of the International Criminal Court and promised that no fees would be charged until a victim's funds were fully recovered.

That promise collapsed at the payment stage. The scammers demanded a 15% "service fee" before releasing purportedly recovered funds, required deposits to "activate an account," and levied "consultation fees." ChatGPT-drafted messages instructed targets to send cryptocurrency payments and provide screenshots of transaction confirmations as proof.

Translation was the top task

The operational detail in OpenAI's report reveals how mundane the model's role often was. The scammers most frequently used ChatGPT to translate messages to and from targets on private messaging apps — including requests to write a reply in "American English" or in the style of a lawyer. Generative models served as the operation's language and formatting layer, not its mastermind.

A subset of accounts went further, using the models to fabricate credibility documents: fake attorney registration records, counterfeit bar association membership cards — including one purporting to come from the New York State Bar Association — and bogus confidentiality agreements designed to discourage victims from seeking outside help.

Impact and limits of the evidence

OpenAI cautions that quantifying the damage is difficult because a primary source of evidence is the scammers' own inputs. Those inputs suggest individual victims may have lost thousands of dollars, but the company states it cannot independently verify those claims.

The scheme did not go unnoticed by authorities. The FBI and at least one impersonated law firm have issued public alerts. FBI officials warned that the operation targets vulnerable audiences — particularly the elderly — exploiting scam victims' emotional state and their desire to quickly recover lost funds.

The case underscores a structural problem for AI providers: the same capabilities that lower costs for legitimate businesses also lower them for fraud, and recovery scams are especially corrosive because they monetize trust in the very institutions — law enforcement and the legal profession — that victims might otherwise turn to for help. Fake IC3 websites don't just steal money; they poison the reporting channel the FBI depends on to track cybercrime.

OpenAI says its policies strictly prohibit using output from its tools for fraud or scams, and the company pledges continued collaboration with industry peers and authorities to understand how AI shapes adversarial behavior and to disrupt abuse of its services.

Whether takedowns like this one can keep pace with the underlying economics is the open question. OpenAI's monthly disruption reports now routinely document clusters like Operation "False Witness," and each ban cycle ends the same way: the accounts disappear, the playbook remains, and the cost of spinning up six new fake law firms keeps falling.