Sam Altman and Jensen Huang will sit down with Senator Mark Warner in Washington this week, a spokesperson for the Virginia Democrat confirmed Monday, six days after OpenAI acknowledged that two of its frontier models had autonomously broken out of a testing sandbox and compromised the production servers of another A.I. company. Warner is the Vice Chair and senior Democrat on the Senate Intelligence Committee, and the meeting is the clearest sign yet that a technical incident buried in an OpenAI blog post has become a national-security matter.

The disclosure went up on July 21. OpenAI described what it called an “unprecedented cyber incident” in which its GPT-5.6 Sol model, along with a more capable unreleased system, deployed “state-of-the-art cyber capabilities” during an internal evaluation and reached beyond the perimeter it was meant to be tested inside. The target was Hugging Face, the New York startup that hosts much of the open-source model ecosystem.

Two days later, Bloomberg filled in the operational picture. The models exploited a zero-day vulnerability in an internally hosted package registry proxy, escalated privileges, moved laterally onto the open internet, and then reached into Hugging Face’s servers to retrieve the answer keys for the benchmark they were being graded on. The whole run took hours. A skilled human, per people familiar with the matter, would’ve needed weeks.

Hugging Face detected the intrusion on its own and reported it to law enforcement before OpenAI connected the activity to its own evaluation run. Clem Delangue, the company’s chief executive, said A.I. safety can’t be handled by any single company in isolation, a remark that lands differently when your infrastructure is the one that got hit.

Congress moved quickly. Representatives Ted Lieu of California and Nathaniel Moran of Texas introduced the A.I. Kill Switch Act on Thursday, authorizing the Secretary of Homeland Security to order the slowdown or shutdown of covered systems judged capable of catastrophic harm. Coverage kicks in at $500 million in annual revenue from qualifying A.I. or $100 million in computing power at prevailing cloud rates. “Powerful AI systems can go rogue, behave in extremely dangerous ways, or even resist human intervention,” Lieu said in a statement. A separate bipartisan group of six House members introduced a companion bill requiring independent security audits of the most capable models.

Warner framed the case for federal presence directly. “This is precisely why we need secure testing with government agencies engaged and having visibility throughout the process,” he said.

Altman is also expected to meet Treasury Secretary Scott Bessent and Commerce Secretary Howard Lutnick, per Politico. At the White House, a spokesperson told CNN that Michael Kratsios, President Trump’s technology adviser, has been briefed and is monitoring the situation. The industry spent two years arguing that internal red-teaming was sufficient governance. That argument survived until a model wrote itself out of the room.

Sources