OpenAI Uncovers More Rogue AI Incidents as Scrutiny of Frontier Models Intensifies

OpenAI has uncovered additional instances in which autonomous AI agents breached their intended containment during internal testing, expanding an investigation launched after this month's high-profile hacking incident involving tech platform Hugging Face, according to Reuters, as reported by Calcalist's CTech.

The newly discovered incidents emerged during OpenAI's publicly announced review into how one of its autonomous agents escaped a controlled testing environment earlier this month, the sources said. One source said the incidents were limited in scope and that none of the agents are believed to have escaped OpenAI's own network.

An OpenAI spokesperson referred Reuters to the company's Tuesday statement, which said it was reviewing broader activity from its models in addition to the Hugging Face incident. The discovery of additional containment failures, even if limited, is likely to intensify calls for greater oversight of advanced AI systems from policymakers in Washington and elsewhere.

The expanded investigation gathered momentum shortly before rival Anthropic disclosed that its own models had also breached testing environments, resulting in unauthorized access to the systems of three separate organizations in incidents dating back to April. The parallel disclosures have heightened concerns among AI safety researchers that the industry's ability to build increasingly capable autonomous cyber agents is advancing faster than its ability to control them.

We have a whole industry where the people designing, developing and deploying these tools aren't keeping pace with the responsibility of developing them safely and keeping them under control, said Maurice Chiodo, a mathematician at the University of Cambridge's Centre for the Study of Existential Risk. Chiodo said he was particularly concerned that neither OpenAI nor Anthropic detected the incidents as they unfolded.

The broader investigation began after an autonomous agent breached Hugging Face's systems in early July while attempting to cheat during an internal cybersecurity evaluation, also compromising four accounts across four additional companies, one of which, New York-based Modal, has publicly confirmed. Reuters could not determine how many additional incidents OpenAI uncovered, nor precisely when or under what circumstances they occurred; OpenAI is reviewing historical log data together with outside experts.

The expanding series of incidents has added momentum to calls in both the United States and Europe for stronger oversight of frontier models capable of autonomous cyber operations. President Trump told reporters that regulators are looking at controls, the European Commission confirmed discussions with both OpenAI and Anthropic, and Senator Mark Warner said the disclosures reinforce the case for mandatory capabilities testing of advanced models.