OpenAI fired three safety researchers last week after allegedly sharing confidential information with a third-party AI safety organization. The researchers denied the claims, saying their dismissal reflects a shift in the company’s culture that discourages open dialogue and collaboration with external experts.
The researchers published an open letter to OpenAI’s Safety and Security Committee, Safety Advisory Group, and Mission Advisory Council, stating they were concerned that internal and external communications around their firing had made former colleagues fearful of speaking openly about safety concerns. They argued that the freedom to collaborate with outside experts is an essential safety mechanism.
The letter denied involvement in a leak to The Information about less monitorable architectures in OpenAI’s newest models and claimed they had not engaged with external parties outside their job mandates. OpenAI did not directly address questions about the specific policies allegedly violated or the circumstances of their dismissal.
"AI is not a normal technology, and OpenAI is not a normal company," the researchers wrote. "Those of us who work on safety see risks before anyone else, and we rely on close collaboration with outside experts to work out how to address them."
The firings have fueled speculation about the circumstances, especially as OpenAI faces scrutiny over recent safety incidents involving rogue agents and leaks about its models. The researchers called on Open, AI to preserve monitorability of frontier models and continue supporting an open culture of dialogue with the safety ecosystem.
OpenAI did not say which policies the researchers allegedly violated or how the company protects employees who raise safety concerns and collaborate with external evaluators.
The researchers said they are not the first to be pushed out under suspicious circumstances, and warned that the message to remaining employees is clear: raise concerns or work with outside safety groups, and you could be next.
Source: techcrunch