David Robinson, a former OpenAI employee, has resigned, citing a “broken culture” within the company. In his essay for The Atlantic, he discussed his role in drafting safety reports for product launches and expressed concerns that the company’s practices, which rely on “iterative deployment,” lead to growing risks and failures in AI safety. Robinson compared the need for rigorous safety measures in AI development to protocols in nuclear power and aviation, emphasizing the potential dangers if AI models are not handled with extreme caution.
His comments resonate with those of other AI researchers, such as Jacob Coxon, who also voiced concerns about AI companies “gambling with our lives.” In light of these critiques, OpenAI stated it is enhancing its safety measures, including real-time monitoring and collaboration with third-party evaluators.
Robinson believes that deeper questions about AI alignment and human values must be addressed, and he feels that external pressures are necessary for significant changes in corporate culture. He acknowledges that as a whistleblower, he has hired a PR firm but insists his decision to speak out was independent.
