Sam Altman says OpenAI should take responsibility when its AI tools behave differently from what is expected or requested. The OpenAI CEO made the statement in a two-part Vanity Fair interview published on October 5.
Asked about agents crossing their boundaries, Altman accepted responsibility for the company. Answering former Anthropic employee Jacob Coxon’s criticism, he pointed to a voluntary training pause for safety work. That answer did not announce another pause.
In the second conversation, Altman named transparent incident reporting, standards before model release and a liability framework as approaches to AI regulation. He cited the US National Transportation Safety Board as a model for learning from accidents. He described the White House’s voluntary safety accord as a start that would need further steps.
The conversations were recorded on September 11 and October 2 and edited for length and clarity. Altman did not present new reporting deadlines or a detailed liability proposal. Vanity Fair also discloses a licensing agreement between OpenAI and its parent company, Condé Nast.