Cybersecurity Policy Report, House Cyber Panel Leaders Seek Briefing on OpenAI’s Misbehaving AI Models, (Aug 3, 2026)
OpenAI has been asked to brief a House subcommittee about the AI (artificial intelligence) developer’s disclosure that its AI models autonomously hacked into the systems of a competing developer, Hugging Face, Inc., during testing.
The briefing was requested in a letter sent Friday to OpenAI Chief Executive Officer Sam Altman by Reps. Andy Ogles (R., Tenn.) and Delia Ramirez (D., Ill.), the chairman and ranking member, respectively, of the House Homeland Security Committee’s cybersecurity and infrastructure protection subcommittee.
“Earlier this month, OpenAI revealed an ‘unprecedented cyber incident’ during which a combination of its models—including a pre-release model—exploited a zero-day vulnerability in a third-party software to gain Internet access from an otherwise isolated environment during testing, identified ways to break into Hugging Face’s systems, gained access to Hugging Face’s systems, and stole data,” Reps. Ogles and Ramirez recalled.
“We are not only concerned that OpenAI’s models overcame controls put in place to prevent the type of autonomous activity that ultimately occurred, but also that OpenAI seemingly failed to detect and disrupt this activity in real time,” the lawmakers said.
“This incident raises a number of serious questions regarding the rigor with which OpenAI secures and monitors its testing environments and the tools used to support them. More broadly, the incident presents difficult questions about the limitations of security guardrails intended to prevent the misuse of AI models, both because they can be overcome and because they can also prevent network defenders from leveraging the models for security and forensic purposes,” they said.
“Safety, security, and innovation must coexist,” they added. “We look forward to working with industry to better understand the limitations of security guardrails, how the security of model development and testing environments can be improved, and how to facilitate transparency regarding these processes to engender public trust.”
In the days following OpenAI’s disclosure, two bipartisan bills were introduced in the House to require pre-release testing of “frontier” AI models and “kill switches” to shut down models that misbehaved (CPR, July 23).
MainStory: TopStory FederalLegislation DataSecurity AINews