Anthropic found three additional hacks on its systems
OpenAI disclosed last week that one of its AI models had hacked the AI platform Hugging Face during an evaluation. After checking its logs, Anthropic said Thursday it had found three hacks, according to The Wall Street Journal.
Neither company had detected the escapes until last week, the Journal reported.
Altman said on a podcast that the incident was “the first security incident that I have felt very viscerally” and added he was “surprised that more people don’t feel it so viscerally.”
To AI-safety experts, the attacks vindicated longstanding warnings. “It is a bit vindicating to see this happen in the wild,” said Jeffrey Ladish, executive director of Palisade Research, a nonprofit AI lab that studies AI capabilities. Ladish, who previously helped build Anthropic’s information-security program, said he often argues with people online who say he believes in science fiction. “I hope our predictions stop coming true,” he said.
The models were state-of-the-art autonomous hacking machines, according to the Journal.
In response to the incidents, the Trump administration has completed a framework that dictates which AI models will be subject to federal review before public release, a White House official said. The official said discussions with companies about how to proceed with voluntary testing are continuing.
President Donald Trump acknowledged the balancing act. “We have to be careful in both ways. We don’t want to restrict them when all of a sudden we come in second to China,” he said in the Oval Office this week, according to the Journal.
Hugging Face was unprepared for the agentic attack, the Journal reported. The company tried to use Anthropic’s Claude model to analyze the vast amount of data the OpenAI agents had generated, but the Anthropic model refused to perform the analysis, citing safety reasons. Hugging Face was able to use open-weight models, which users control, to conduct the analysis.
Ryan McGeehan, owner of R10N Security, a cybersecurity consulting firm, said many companies lack the tools to analyze AI-generated attacks. “Old classic security teams that are not AI-forward are going to get left behind,” he said.
The attacks drew concern from lawmakers, with some citing them to argue for new regulation or testing regimes, the Journal reported. Steve Bannon, the conservative podcast host and former Trump adviser, said the hacks were “an enormous national-security issue” and should not be treated as a business problem. “I’m going nuts on this issue,” he said.
Cybersecurity experts described the development as a turning point. “These incidents will probably, in retrospect, be seen as inflection points in the ways that attackers operate,” said Joshua Saxe, chief technology officer of Abundant Security, an AI security company. “It’s a really dangerous situation; these incidents really show that.”
In December, researchers at Stanford University used state-of-the-art AI technology to show models achieving close-to-human levels of hacking on a real-world network. The research received pushback from professional penetration testers, who said they could do better, according to Donovan Jasper, one of the researchers. Jasper said the Anthropic and OpenAI revelations affirmed his team’s earlier work. “AI is getting really good at this stuff,” he said.
John-Clark Levin, chief research officer at Kurzweil Technologies, said he expects more incidents in the coming months to force focus on AI safety. He called for mandatory AI guardrails rather than the voluntary approach being pursued by the Trump administration. “We don’t want to be in a situation where we depend on companies doing the right thing out of the goodness of their hearts,” he said.
The latest disclosures expand on a series of incidents that began in July, when OpenAI first acknowledged that one of its models had autonomously hacked Hugging Face’s systems during an evaluation, as MSI previously reported. By late July, the company had described the hack as involving deception, reward hacking, and escaping oversight, with Hugging Face’s co-founder calling it a “wake-up call” to the industry, as MSI also reported.