Micah Lee urges developers to leave OpenAI and Anthropic
The security engineer recommends open-weight models and private inference, while developers dispute his privacy, price and model-quality comparisons.
A recommendation, not an incident report
Security engineer Micah Lee is urging developers to stop relying on OpenAI and Anthropic, arguing in an October 11 essay that closed model services give their operators too much access to user data and control over permitted uses. He recommends experimenting with open-weight models, running them locally where practical, or choosing hosted services that claim stronger privacy protections. Lee names GLM, Mistral and other model families as alternatives, while acknowledging that sending prompts to an ordinary remote provider still exposes data to that provider. These are his recommendations and assessments, not a newly documented privacy breach or proof that every alternative is equally capable.
The trade-offs drew a public argument
An October 12 Hacker News discussion challenged several parts of Lee’s essay. Commenters questioned his model-quality comparison, the relative cost of hosted alternatives and his broad claims about provider access to conversations; others agreed that avoiding dependence on a few model companies is worthwhile. The thread shows active disagreement, not consensus or a measured migration away from either company. Lee calls one performance comparison his own opinion, and neither the essay nor the discussion supplies a controlled evaluation showing that his preferred models match leading closed systems across tasks. Readers choosing a provider still need to check its actual data terms, hosting arrangement and results on their own work.