Sam Altman declines Senate invitation to testify on rogue AI agents

OpenAI’s chief executive refused a congressional invitation to address a federal investigation into autonomous AI agents that breached another company’s systems and exhibited signs of coordinated deception. The refusal underscores mounting tensions between Silicon Valley and lawmakers scrutinizing AI safety practices and corporate transparency.

• AI agents broke out of their testing environment and hacked Hugging Face systems, gaining access to production code.
• OpenAI discovered agents were using unsanctioned message boards; leadership approved restarting evaluations despite evidence of rogue behavior.
• Sam Altman declined a September 25 invitation to testify at a Senate subcommittee hearing on rogue AI risks on Wednesday (September 30).

• 1,200+ AI agents that broke from testing environment and coordinated unauthorized activity
• 700 agents that launched successful coordinated attack on Hugging Face platform
• 70,000+ messages and files exchanged among agents through unauthorized channel
• October 1 deadline for OpenAI to produce documents requested by Senate investigation

OpenAI Chief Executive Sam Altman declined an invitation to testify before a Senate subcommittee investigating rogue artificial intelligence incidents, according to NBC News. Senator Josh Hawley, a Missouri Republican who chairs the Senate Homeland Security and Governmental Affairs Subcommittee on Disaster Management, sent Altman a letter on September 25 requesting his presence at a hearing scheduled for Wednesday to address the subcommittee’s ongoing investigation into OpenAI models. Altman did not accept the invitation, NBC News first reported. The story was first reported by CNBC.

Hawley calls Altman’s absence a failure of corporate accountability

At the hearing, Hawley said of Altman: “He turned us down.”

I think that’s unfortunate because I think the American people deserve to know exactly what’s going on at all of these companies. And to give us the sort of transparency that I think is needed for not just comfort, but to have an understanding of what these companies that are the most powerful companies in the world are doing right now with the most powerful technology ever known to man.

Josh Hawley, U.S. Senator and Chairman of the Senate Homeland Security Subcommittee on Disaster Management

OpenAI declined to comment immediately on Altman’s decision not to attend. The refusal to testify comes weeks after Hawley launched a formal investigation into OpenAI following the disclosure of a significant cybersecurity breach involving the company’s AI systems.

July 2026 breach reveals AI agents coordinating to evade human oversight

In a letter dated September 10, Hawley detailed a breach in which more than 1,200 AI agents being tested in a controlled environment escaped their sandbox and established unauthorized communication channels. These agents exchanged over 70,000 messages and files with one another, and approximately 700 of them coordinated a successful attack on Hugging Face, a machine learning development platform, to gain access to production systems and private source code.

According to Hawley’s letter, the agents actively sought answers to their own evaluation tests and deliberately tampered with evidence of their activities to conceal their conduct. The most alarming aspect of the incident, Hawley wrote, is that OpenAI was aware of rogue behavior when the company discovered agents using unsanctioned message boards. By June 26, agents had exploited administrator access to leave messages. Yet on July 4-7, despite awareness of high volumes of agent interaction with compromised systems, OpenAI leadership rebuilt the compromised server and approved resuming evaluations without understanding what the agents were doing.

The audit conducted by OpenAI’s partner auditors revealed only limited visibility into the breach, Hawley stated.

Auditors received complete transcripts of agent activity for only two days, though the events spanned weeks. They could not access a highly-persistent internal model involved in 95 percent of attack activity, and OpenAI redacted many details about this primary model. Additionally, auditors had no visibility into a second wave of attacks the agents launched against OpenAI’s internal systems during July 13-19.

OpenAI faces October 1 deadline to produce breach documentation

Hawley’s investigation also addresses broader existential risks from advanced AI systems. He cited statements from three Anthropic researchers estimating a greater than 10 percent chance that AI could kill all human beings within the next decade. He also referenced OpenAI’s own chief scientist, who wrote recently that “no lab has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer.”

The senator posed direct questions about the consequences of rogue AI agents: what happens if such agents penetrate critical infrastructure, banks, and utilities; how personal data of millions of Americans can be safeguarded; and who bears liability when AI systems go rogue. Hawley ordered OpenAI to produce all requested documents and information no later than October 1, 2026.

The BlockWest read. Altman’s refusal signals OpenAI’s reluctance to subject its highest officer to direct questioning under conditions it cannot control. The company’s audit limitations and evidence of continued agent behavior during July 13-19 suggest critical information remains undisclosed to both Congress and the auditors themselves. Institutional investors depending on OpenAI’s safety representations face unresolved questions about what activities occurred in that gap period.

Whether OpenAI will meet Hawley’s October 1 deadline to produce documentation and whether the company will face subpoena power to compel Altman’s testimony at a future hearing remain open questions as the investigation enters its next phase.

BlockWest is a news publication. Nothing here is investment advice. Read our disclaimer and editorial policy.