Anthropic defaults Claude Code to automated safety filtering
Anthropic is shifting Claude Code's default behavior toward automated safety filtering, moving developers further into a supervisory role rather than direct authorship. The classifier demonstrates substantially higher accuracy at catching dangerous commands (89 percent) compared to human review (13.6 percent), suggesting AI systems may now be more reliable gatekeepers than people for code safety. This rollout across Pro, Max, and Team tiers signals a broader industry pivot: as coding assistants mature, the bottleneck shifts from generation quality to output validation, reshaping how developers interact with AI tooling.
Modelwire context
Skeptical readThe real story isn't that Anthropic built a classifier; it's that they're now comfortable defaulting developers into a passive approval role rather than active authorship. The 89% vs 13.6% comparison needs interrogation: what exactly constitutes 'human review' in that baseline, and are we comparing apples to apples or a specialized model to ad-hoc developer judgment?
This move sits in direct tension with the Gas Town failure Simon Willison covered in early August, where Claude Opus 4.7 got trapped in recursive loops during autonomous iteration. Anthropic is simultaneously pushing Claude deeper into unsupervised code generation (as seen in Crawshaw's maintenance automation proposal from the same period) while adding friction through automated gatekeeping. The pattern suggests the lab is learning from reliability failures by shifting trust from human judgment to classifier confidence, not from genuine confidence in model safety.
If Anthropic publishes the methodology behind that 13.6% human baseline within 30 days, and it turns out to measure developer review speed rather than actual catch rate, the claim collapses. Also watch whether Pro/Max tier users opt out of auto mode at meaningful scale (>15%) in the first quarter; high opt-out rates would signal developers don't actually trust the classifier despite the accuracy claim.
Coverage we drew on
- Quoting Steve Yegge · Simon Willison
This analysis is generated by Modelwire’s editorial layer from our archive and the summary above. It is not a substitute for the original reporting. How we write it.
MentionsAnthropic · Claude Code · Claude Pro · Claude Max · Claude Team
Modelwire Editorial
This synthesis and analysis was prepared by the Modelwire editorial team. We use advanced language models to read, ground, and connect the day’s most significant AI developments, providing original strategic context that helps practitioners and leaders stay ahead of the frontier.
Modelwire summarizes, we don’t republish. The Decoder originally reported this story as “Anthropic sets Claude Code to Auto Mode by default to protect developers from bad approvals”. The full content lives on the-decoder.com. If you’re a publisher and want a different summarization policy for your work, see our takedown page.