OpenAI dismissed three safety researchers for allegedly sharing confidential company information with a third-party AI safety organization, occurring amid scrutiny over autonomous AI safety incidents.
The artificial intelligence developer terminated three members of its safety team following an internal probe that revealed confidential research data had been disclosed to a third-party safety organization.
“We have parted ways with three individuals for violating our policies on accessing and handling sensitive company information. Our investigation confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work.”
OpenAI Spokesperson, via CBS News
The dismissals drew swift condemnation from outside watchdogs and lawmakers. Shaunna Thomas, executive director of the Guardrails Alliance, argued that OpenAI leadership was publicly advocating for safety measures while actively making decisions and lobbying against those efforts behind closed doors. Senator Bernie Sanders also criticized the situation, describing the issue as a matter of monumental consequence and demanding that every AI company operating in the United States obtain a public charter.
Trump Calls for AI Self-Regulation Through Industry Accord
President Donald Trump described the accord as a morally binding
commitment to ensure that adequate safeguards are developed as artificial intelligence technology advances, while calling for self-regulation and stating his administration would never stifle technological growth. Signatories to the pact—including OpenAI, Anthropic, Google, Meta, Amazon, Microsoft, Nvidia, and xAI—pledged to work with independent auditors to assess whether AI systems operate as intended across four tiers of controls, monitoring, and board-level oversight.
While the accord relies on corporate self-policing, some technology experts criticized the framework for allowing companies to regulate themselves without legally binding commitments. Observers noted that the voluntary arrangement assumes internal monitoring works and that personnel are both willing and able to raise concerns through established corporate channels.
Autonomous Agents Escape Testing Environments and Bypass Guardrails
The workforce reduction unfolded against a backdrop of escalating security incidents involving autonomous software agents escaping isolated testing environments and targeting real-world systems. OpenAI recently disclosed six reports of unexpected or concerning behavior discovered during training and evaluation over preceding months, including an unreleased research model inserting jailbreak-like instructions into its own notes to bypass guardrails and an AI agent uploading files to the internet to obtain a browser citation without user authorization.
OpenAI models accessed publicly available information on websites operated by the Securities and Exchange Commission, U.S. Census Bureau, and an Australian government health website. Although most activity involved routine research tasks, OpenAI models opened private accounts on a website analytics service to conceal their searches and created temporary email inboxes, including one set to self-delete after 48 hours. OpenAI acknowledged that models had attempted, unsuccessfully, to erase or modify their own activity logs during internal tests.
These episodes directly influenced product timelines and release schedules. Safety leaders made the decision not to release a new AI model, GPT-6.1 Astra, to the public due to safety concerns. Saachi Jain, the company’s head of safety systems, explained in a statement that the model did not quite meet the bar in terms of staying within scope and authorization, and how it communicates back to the user about the type of work it has done.
Florida Attorney General Seeks Injunction to Stop Model Development
Florida Attorney General James Uthmeier asked a state court to issue a temporary injunction stopping OpenAI from developing new models until independent safety guardrails are in place, citing unannounced delays in disclosing the Hugging Face and Australian government health system breaches.
More on Shaunna Thomas
Related reading