OpenAI Firings Spark Concerns Over AI Safety Discourse

Three safety researchers, recently dismissed by OpenAI, have issued an open letter contesting the company’s narrative regarding their terminations. They warned that the situation might deter employees from discussing AI risks candidly.

Dismissed Researchers Address Concerns

Tomek Korbak, Jasmine Wang, and Mikita Balesni directed their letter to OpenAI’s Safety and Security Committee, Safety Advisory Group, and Mission Advisory Council. They expressed worry that communication surrounding their dismissals had instilled fear in former colleagues, preventing them from speaking out.

According to the trio, safety researchers need freedom to challenge decisions and engage with external experts. They emphasized that AI, and OpenAI as a company, operate outside conventional norms.

OpenAI’s Stance

OpenAI has denied that the firings related to voicing safety concerns, claiming an investigation revealed violations of policies regarding sensitive data.

The letter titled “OpenAI cannot make AI safe on its own” publicly warns of the impact these dismissals may have. The researchers stated, “Internal and external communications around our firing have made our former colleagues afraid to speak.” Previously, OpenAI encouraged raising safety concerns, but following these events, staff are uncertain about their position.

Disputing OpenAI’s Claims

Korbak, Wang, and Balesni refuted claims about the reasons behind their departures. They denied being the leak source regarding AI architecture developments and rejected claims of unauthorized communications with outside organizations.

Specific Allegations Against Korbak

Korbak alleged he was amiss for his communication with Model Evaluation and Threat Research (METR), although it was part of his job. He cited raising concerns about monitoring AI agent reasoning as a potential reason for his dismissal.

Balesni’s Challenges

Balesni speculated their terminations occurred because they prioritized safety over OpenAI’s short-term goals. During an exit call, he was accused of excessive communication with external safety groups but denied sharing proprietary information.

Wang’s Account

Wang challenged OpenAI’s reason for her termination—accessing an executive’s email. She clarified the access was for recruiting, and she requested its removal when no longer needed.

Recommendations to OpenAI

The researchers urged OpenAI and other AI companies to continue independent safety assessments and maintain external collaboration for ensuring AI safety. They warned against weakening ties with external organizations overseeing powerful systems.

They also emphasized the need for transparent discussions with independent experts to preserve model monitorability.

OpenAI’s Response

OpenAI responded firmly, stating the dismissals followed a thorough investigation uncovering significant policy breaches. It stressed those decisions were not about silencing safety concerns.

OpenAI remains committed to independent safety evaluations. They are finalizing contracts with third-party assessors and acknowledged the importance of collaboration in preserving AI monitorability.

They praised the contributions made by the dismissed researchers while insisting that trust remains vital to their culture. OpenAI plans to continue encouraging open debate as part of AI development.

Why This Matters

The conflict underscores challenges faced by AI developers in balancing confidentiality with public safety assessments. The former researchers believe relationships with independent evaluators weaken if employees fear job security risks for speaking candidly.

Their letter concludes by emphasizing the critical role researchers play within organizations in ensuring safe AI development.