Claude Code's auto mode quietly uses a second safety classifier model, sparking Max-tier transparency complaints
tomekkorbak · x · 2026-09-16
Developers discovered that Claude Code's auto permission mode runs a second model as a safety classifier reviewing the main agent's actions — an error message revealed its existence.
- Official docs confirm: in auto mode (default on Pro/Max/Team plans), a classifier model reviews actions instead of the user; manual mode requires per-action approval.
- User michptaszynski objected that paying for Opus Max means expecting max effort on everything, not silent delegation to a weaker model.
- tomekkorbak argued a cheaper model makes sense for low-difficulty, low-latency safety checks — noting that users not noticing the classifier proves it's good enough.
The exchange lifts the curtain on Claude Code's multi-model architecture and raises transparency questions about how paid tiers are actually served.
Related event: Hidden safety classifier model in Claude Code sparks user debate(3 posts)→
More from coding & agent
- Microsoft's MemSteer turns unlabeled user histories into agent memory supervision — MicrosoftResearch · 2026-11-02
- She fired herself from her own Instagram: AI agent runs the account, views up 82.3% — heyshrutimishra · 2026-11-02
- Sourcegraph: AI writes code faster than teams can review it, driving more outages — AI Engineer · 2026-11-02
- Tsinghua & Microsoft's HarnessMM cuts agent token use by up to 87.93% while raising success rate — rohanpaul_ai · 2026-11-02
- GitHub launches Agent HQ: Anthropic, OpenAI, Google coding agents in Copilot — emmanuelvivier · 2026-11-02
- DeepSeek open-sources Super-Agent, a coding agent that reads team chats to minimize human input — emmanuelvivier · 2026-11-02