Ethics and safety·October 10, 2026, 05:41

Former Anthropic employee: AI companies aren't secure enough

AI-generated and checked against the sources listed below.

David Robinson, who recently left Anthropic, says in an interview with Ezra Klein that OpenAI and other AI companies rely on security practices that are too weak, even as their systems become more dangerous.

Journalist Ezra Klein has interviewed David Robinson, who recently left the AI company Anthropic. The conversation focused on how safely AI is being developed.

Robinson says OpenAI and its competitors are not operating securely enough. According to him, the industry still uses security practices suited to small startups, even though they are building systems that are becoming more and more dangerous.

Failures at multiple companies

Robinson points to security incidents that have been documented at OpenAI, Hugging Face and Anthropic. Among other things, a security mechanism at Anthropic was configured incorrectly.

Comparison with nuclear power

Robinson believes that if control over AI systems is lost, the damage could be greater than that of a meltdown at a nuclear power plant. He also says the companies' internal controls are far from the standards required at nuclear power plants.

For readers, this means that someone with inside knowledge from one of the leading AI companies believes security is not keeping pace as the technology grows more powerful.