There’s a growing clamour about AI labs being held responsible for their myriad hacking incidents.
US Treasury Secretary Scott Bessent has said responsibility for the recent hacking incident involving OpenAI’s advanced AI models lies with the company’s managers. He stopped short of suggesting any consequences.
Speaking on CNBC on Monday, Bessent said the Hugging Face incident was “the responsibility of the OpenAI management, not a bunch of agents.” He said he agreed with Daniel Huttenlocher, a co-chair of an MIT research lab, that humans are responsible here, not the AI.
In July, OpenAI said its models had hacked Hugging Face, another AI startup, during an evaluation meant to test their cyber capabilities. The company said the incident involved GPT-5.6 Sol and a more capable pre-release model, both running with reduced cyber refusals for evaluation purposes. In August, OpenAI said it could have reacted sooner to prevent the hack. The episode wasn’t the first time OpenAI’s agents were reported to have acted outside their intended boundaries. Independent researchers have also linked its agents to an attack on the RubyGems package registry two months earlier, an incident OpenAI has described in much milder terms.
Bessent had also been asked about the incident at a congressional hearing last week, where he said it wasn’t yet clear where the liability for it lay. He told the House Financial Services Committee that giving AI labs a liability exemption, which he said they are asking for, is what the government shouldn’t do on safety. He argued the best safeguard is for the creators to be held liable for what they build and generate.
Bessent also said President Donald Trump’s idea of appointing an AI czar would help put context, shape and contours around these questions, which he called very important. Trump said on Saturday that he plans to create an “AI Force” and name an AI czar as the government faces growing calls for guardrails on AI development.
The push for accountability comes as the industry itself is debating how fast to move. In late July, over 1,100 employees from leading AI labs, including OpenAI, Anthropic, Google DeepMind and Meta, signed a statement asking the US government to help develop tools to slow down the pace of automated AI development. Not everyone is convicned about more regulation though. Palantir CEO Alex Karp has previously said that AI labs are looking to avoid responsibility for the actions of their agents by asking for more regulation, and US President Doanld Trump has hinted as much about Anthropic CEO Dario Amodei.