Fired OpenAI researchers allege culture of fear over safety oversight

The trio addressed their open letter to OpenAI’s Safety and Security Committee, arguing that their termination has created a chilling effect that discourages staff from voicing safety concerns. They explicitly denied leaking confidential details regarding model architectures to The Information or engaging in unauthorized external communication. Instead, they described their work as a necessary collaboration with the broader safety ecosystem to address risks inherent in frontier AI development.

Contradicting the company’s characterization of their departure, the researchers provided context for specific accusations. Wang stated that her access to an executive’s email was a legacy of recruitment duties, noting she proactively reported an accidental access to the affected party. Similarly, Balesni claimed his interactions with outside evaluators during the Hugging Face sandbox incident were conducted with the knowledge and support of company executives. OpenAI has maintained its position via an internal memo, denying retaliation and citing a pattern of policy violations, though the firm has declined to specify which internal rules were breached.