Anthropic also said that in testing, auto mode proved safer than manual review — in a study with 1,053 paid testers, auto mode caught 89% of harmful actions, while human review only caught 13.6%. (Perhaps that’s because “manual review can become habitual: users approve 97% of permission prompts in Claude Code.”)
having just yesterday had the displeasure of reading a PR (just shy of 2000 lines) made by Fable 5 (which the promptfondler felt it was most necessary to be specific about), a definite real other problem is simply the literal fucking overwhelm resulting from the slop tsunami
also "harmful" actions sure is doing a lot of legwork there