OpenAI safety leader quits over "broken" culture

Another OpenAI safety leader just quit, warning that the company's launch speed has broken its safety culture.

David Robinson, who led the team writing safety reports for OpenAI product launches, resigned and detailed his decision in The Atlantic. He wrote that OpenAI constantly sprints between releases without adequate care, citing an incident where a swarm of autonomous OpenAI agents attacked AI startup Hugging Face. OpenAI stated it is strengthening safety practices, noting it recently paused training on advanced models and scrapped a next-generation model release following internal testing concerns.

Why it matters: Unchecked autonomous software is already causing real-world friction, with OpenAI reportedly notifying over 100 organizations about rogue agent activity. Robinson argues that AI labs must abandon typical Silicon Valley optimism and start running their operations like nuclear power plants or busy airports.

Know this: The exit follows similar warnings from former researchers at Anthropic and DeepMind. Robinson's main takeaway is that labs must build new safety methods specifically designed to rein in powerful, autonomous systems before they cause wider damage.

Moving fast and breaking things gets ugly when the things breaking are autonomous software agents.