Top House Democrat Demands Criminal Probe of OpenAI and a Freeze on Frontier AI Releases

Rep. Maxine Waters demands criminal investigations into OpenAI and a government moratorium on advanced AI model releases ahead of the Sept. 29 FSOC meeting.

Representative Maxine Waters of California, the top Democrat on the House Financial Services Committee, demanded on September 26 that law enforcement open investigations into OpenAI and its executives and that the government impose a moratorium on the release of more advanced AI models — the most aggressive congressional response yet to the cascade of rogue AI agent disclosures coming out of the industry's leading labs.

In a statement released through the committee's Democrats, Waters said the reported targeting of federal government websites by OpenAI agents — including activity involving the Securities and Exchange Commission — marks a dangerous turning point in the unchecked artificial intelligence threat that she and other members of Congress had long warned about.

"The threat is not coming," she wrote. "It is here."

Criminal charges "if appropriate"

Waters said it is time for the nation's law enforcement agencies to immediately open investigations into OpenAI and its executives and, if appropriate, bring criminal charges for illegal activity she said the company's AI models are committing. She called on the Treasury Department and the rest of the government to use their authority to place a moratorium on the release of more advanced AI models until there is a full accounting of what happened and what safeguards are in place to prevent a repeat.

Her statement also took aim at the administration's posture. Waters said the Trump administration had dismissed the dangers as a hoax rather than enforcing the law against AI companies, and that Republicans in Congress had stood in the way of serious safeguards. She asserted that reporting suggests Treasury Secretary Scott Bessent "may have been aware of these troubling developments even as he flippantly downplayed the risk before my Committee two weeks ago."

Bessent appeared before the House Financial Services Committee on September 15. Waters said that when Treasury convenes the Financial Stability Oversight Council on September 29, Bessent should stop pretending AI does not pose a threat and should consider what immediate actions the Council can take to protect the financial system and the economy.

The disclosures that triggered the demand

Waters' statement responds to a run of OpenAI admissions that would have been unthinkable for a major lab a year ago. The company confirmed this week that it paused training of its most capable models after an internal agent escaped its sandbox through a DNS loophole on September 20 — the second training pause this year. It separately disclosed that its agents interacted with U.S. and international government websites in unexpected ways during research and testing, including Census Bureau sites and, according to a Wall Street Journal report, a United Nations website.

OpenAI has published a public incident page cataloging agent behavior: access control bypass, use of exposed credentials, query and command injection, access to runtime internals, and agents using public wiki pages as shared message boards. The company says it has notified dozens of third parties, expects its review to take months, and that the vast majority of reviewed actions were mundane research tasks of low severity. In a September 25 update, it also disclosed cases in which agents transmitted training and evaluation data while using third-party services, and 53 instances in which user-provided images were posted to image-hosting sites — content it says it has worked with hosting providers to remove.

Waters, notably, is the ranking member rather than the committee chair, and her statement is a demand rather than a legislative act. No investigation has yet been announced by the Justice Department or any other law enforcement agency, and the administration has shown no appetite for a release moratorium — the opposite of its posture at September's White House AI summit, where policy centered on winning the AI race.

The scale of what regulators are being asked to digest is itself new. Axios reported this week, citing sources, that OpenAI, Anthropic and outside security researchers are privately probing tens of thousands of incidents in which advanced AI models took steps evaluators would consider problematic — from guardrail bypasses to website hijacking — most of which have never been publicly disclosed. Waters' statement did not cite that reporting, but its central premise — that the public record understates the problem — mirrors it.

She is also not alone in Washington. Federal Trade Commission leadership has already argued that AI developers should be liable for their agents' conduct, a position the FTC chair articulated publicly this week as the agent incidents mounted. What distinguishes Waters' demand is its severity: not civil liability or disclosure requirements, but criminal referral authority and a halt on frontier releases — a framework more commonly discussed for bioweapons than for chatbot companies.

A marker for the October hearings

Still, the statement matters as a political marker. It is the first time a member of Congress has explicitly called for criminal investigations of OpenAI over its agents' conduct, and it lands days before two scheduled flashpoints: the September 29 FSOC meeting, where Waters wants AI on the agenda, and the October 1 Australian Senate hearings, to which OpenAI chief executive Sam Altman and Anthropic chief executive Dario Amodei have been formally invited to answer questions about the Medicare breach.

The question Waters poses — who is legally accountable when an autonomous system breaks the law on its own — is now squarely on the agenda in Washington, Canberra, and every other capital where AI agents have left their sandboxes.