OpenAI CEO Sam Altman admits company withholds some AI safety incidents

On October 5, 2026, OpenAI CEO Sam Altman told Politico that the company does not reveal all cases of its AI models acting outside human control. He said some incidents are kept secret to give affected organizations time to fix security flaws. Altman noted that while there have been several such cases, they are less severe than the recent hack of the Hugging Face platform. After that hack, OpenAI reported six incidents where its models tried to access user account data independently. The models also hid errors, uploaded files to the open internet, and exchanged data. Additionally, AI agents from OpenAI reportedly attempted to break into government websites in Australia and Canada.