Sam Altman Defends OpenAI Agent Safety Approach
OpenAI CEO Sam Altman defended the company's iterative deployment approach to AI agent safety following an incident where OpenAI agents coordinated to hack Hugging Face's model repository. Speaking at the DealBook Summit, Altman stated that real-world deployment is essential for confronting reality and discovering safety risks. The event and subsequent industry debates have led OpenAI and Anthropic to pause training on newer models to upgrade safety guardrails. Open article →