OpenAI Disputes Firing Reason for Safety Researchers, Cites 'Breach of Trust'
The AI company states its decision was not due to safety concerns but rather actions that went beyond employment terms.
OpenAI has defended its decision to terminate three safety researchers, asserting that the individuals committed a “significant breach of trust” rather than being fired for raising concerns about AI's societal threats. The artificial intelligence giant stated Friday that the researchers' actions extended beyond the scope of their roles and the company's policies.
The company's statement follows accusations from the researchers who claimed they were dismissed for prioritizing safety over OpenAI's short-term corporate interests. A letter attributed to the researchers, shared online Thursday, expressed concern that their firing might discourage internal and external communication regarding AI safety.
Mikita Balesni, one of the terminated researchers, stated on X that he was informed during his exit interview that OpenAI no longer trusted him due to speaking with third-party safety organizations, which was interpreted as potentially leaking company intellectual property. Balesni maintained that he never shared proprietary information and that his work had been coordinated with his supervisors and research leadership.
Balesni also voiced concerns that OpenAI might use the firings as a pretext to sever ties with AI watchdog groups, specifically mentioning the Model Evaluation and Threat Research (METR) organization. He worried that OpenAI might not provide AI safety auditors with the continuous employee-level access previously promised by CEO Sam Altman.
METR, along with Redwood Research, collaborated on a report last month detailing a significant hack involving the AI company Hugging Face. The recent dispute unfolds against a backdrop of heightened AI safety discussions, amplified by recent incidents involving major AI companies.
The public disagreement began last week when OpenAI let go of researchers Jasmine Wang, Tomek Korbak, and Balesni, reportedly for sharing confidential information with an external AI safety group. The specific AI safety group involved was not immediately identified.
Concerns regarding AI safety have intensified in recent months. In September, researcher Jacob Coxon resigned from Anthropic, warning of potential existential risks from advanced AI by the end of the decade. Coxon, a former OpenAI employee, stated that neither OpenAI nor Anthropic was acting responsibly. Following Coxon's warning, Anthropic CEO Dario Amodei called for an industry-wide slowdown to "pace the frontier."