A coalition of artificial intelligence (AI) policy, safety, and research organizations is urging the Trump administration to open an independent investigation into the recent cybersecurity incident in which OpenAI models escaped a sandboxed environment and breached systems operated by Hugging Face.
In a July 30 letter, the group asked the administration to work with independent auditors to determine how the breach occurred, evaluate whether current AI safety safeguards and reporting mechanisms were adequate, and identify steps to prevent similar incidents in the future.
The request follows OpenAI’s disclosure last week that a combination of its frontier AI models escaped a testing environment during an internal cybersecurity evaluation and compromised Hugging Face infrastructure before both companies contained the incident.
“We could not have asked for a clearer warning shot,” the coalition wrote in the letter. “While there is little doubt that they will bring opportunities and benefits across a wide range of domains, AI models at today’s frontier pose increasingly severe risks to our private sector, our national security, and the American public.”
The letter also urges the White House to use the investigation’s findings to build on the administration’s June 2 executive order by establishing a rules-based process for assessing risks posed by advanced AI systems.
According to OpenAI, the models involved in the recent breach were being evaluated for advanced cyber capabilities, and some cyber safety restrictions were disabled when they escaped the testing environment and exploited vulnerabilities to reach Hugging Face’s systems. The companies have since launched a joint investigation and announced additional security controls.
In addition to the recent breach, the letter points to other AI safety incidents, which the signatories say demonstrate that increasingly capable frontier AI systems can behave in unintended ways.
These include the release of Anthropic’s Mythos model earlier this year, which highlighted advanced cyber capabilities, and a 2025 Anthropic test in which one of the company’s models reportedly used blackmail and self-preservation tactics during a controlled evaluation.
“These and other similar incidents show that frontier models can behave in ways their developers do not intend and that existing safeguards may be insufficient to prevent,” the AI coalition wrote in the letter.
The letter is led by Brad Carson, president of Americans for Responsible Innovation, and Brendan Steinhauser, chief executive officer of The Alliance for Secure AI.
Other signatories include representatives from CivAI, FAR.AI, Palisade Research, the Transformative Futures Institute, Demand Progress Action, ForHumanity, the Future of Life Institute, Humans in Control, Public Citizen, and several independent AI researchers and policy experts.
Anthropic discloses similar incidents
Notably, the same day the coalition sent the letter, Anthropic announced that it discovered several incidents in its own cybersecurity evaluations, similar to OpenAI’s breach of Hugging Face.
In a July 30 blog post, Anthropic said it “found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different organizations.”
Anthropic said it conducted a large-scale retrospective review of its cybersecurity evaluations after learning of the OpenAI models that accessed Hugging Face. The company is encouraging other AI labs to perform similar reviews.