Dev slams Anthropic's auto mode classifier: stops zero bad actions, blocks legit work
cjimti · x · 2026-10-21
A developer publicly criticized Anthropic's auto mode permission classifier, saying it has been "100% garbage" in practice: it has caught and stopped not a single bad action so far, while constantly blocking legitimate work. In his view, the mode hardly deserves the label "auto."
It's an early negative datapoint on Claude Code's auto permission mode, highlighting reliability concerns for automated permission classification in real engineering workflows.
More from Models
- Dev switches to DeepSeek after Claude Opus refuses basic fantasy violence for a game — QuixiAI · 2026-10-21
- Local LLM for agentic work on 16GB VRAM: gpt-oss-20b refuses too much, what else? — SLTGAMER · 2026-10-21
- OpenAI desktop app degrading: slow session loads and reconnect errors, users report — koltregaskes · 2026-10-21
- One Week After Launch, Meta's Wonder Still Has No Safety Report — ShakeelHashim · 2026-10-21
- Anthropic's unannounced "Fable 5.5" model spotted in public GitHub commit credits — mark_k · 2026-10-21
- 558K evaluations show vision-language models inversely scale on spatial reasoning and STEM — UniversityOfEdinburgh · 2026-10-21