#claude

Anthropic apologized for Claude Fable's invisible guardrails, while benchmarks show mid-tier coding results. Is AI safety degrading developer tools?