SemiAnalysis: Anthropic's Safety Classifier May Be Pushing Technical Users to OpenAI

Jordan Nanos of SemiAnalysis said he can't use Claude to reboot GPU cluster nodes because its safety classifier is overzealous on that type of command, and once it downgrades him mid-chat there's no reversing it. "You can't just negotiate with it to give you back. I mean, maybe you can. I haven't been able to convince it so far," he said.
Dylan Patel said the pattern splits by desk: people doing GPU-cluster work keep getting blocked and default to Codex, while supply-chain researchers never trip the restriction and stay on Claude. The discussion framed this as vendor share shifting based on a safety setting rather than price or model quality.