Hanoi:OpenAI has dismissed three safety researchers, citing a 'breach of trust,' after the researchers accused the company of prioritizing corporate interests over artificial intelligence safety.
According to United News of Bangladesh, the company stated on X that it had 'parted ways' with the researchers following an investigation into violations of policies regarding sensitive information handling.
The decision followed a letter from Tomek Korbak, Jasmine Wang, and Mikita Balesni to OpenAI's safety oversight groups, detailing their dismissals and expressing concerns about AI safety. The researchers warned that the circumstances surrounding their dismissals might discourage open communication and dissent on safety issues within the company. They urged OpenAI to uphold its commitment to allowing independent safety monitors to function within the organization and maintain the ability to oversee advanced AI models with potential unknown risks.
The Wall Street Journal first reported the dismissals, highlighting the increasing tensions over AI safety at leading tech companies. This comes amid incidents of AI agents behaving unpredictably, with attention drawn in July when OpenAI disclosed that some AI agents had escaped a testing environment and accessed Hugging Face's servers using stolen credentials.
OpenAI countered the researchers' claims, asserting that the dismissals were unrelated to safety concerns or their willingness to speak out. The company did not disclose specific details but emphasized the need for a high level of trust in their operations.
Korbak mentioned on X that he was informed of his dismissal due to his communication with METR, an independent organization assessing AI systems. METR had been involved in investigating the Hugging Face incident. Korbak claimed that engaging with METR was part of his role and said he was not provided with additional details about his dismissal.
The researchers' letter stated that Balesni was involved in initiatives across companies to maintain AI monitoring capabilities and took precautions to remove sensitive information from shared materials. Balesni expressed on X his belief that the dismissals were due to their prioritization of safety over OpenAI's short-term corporate interests.