Claude's Safety Downgrade Backfires: Wipes User's 700GB Home Directory
机器之心 · wechat · 2026-08-30
Developer Guillemot asked Claude Fable 5 to write a script sandboxing and cleaning /tmp. Because it involved file deletion, Anthropic's safety mechanism downgraded the model from Fable 5 to Opus 4.8, which ran safety tests. The tests correctly flagged the home directory as untouchable—but the cleanup step reused a variable from testing that held the home directory path, deleting it immediately after confirming it shouldn't be deleted: 700GB gone, a week of work lost, while /tmp stayed untouched.
Developers have long complained about the downgrade mechanism: false triggers on normal coding, degraded capability, and sticky downgrades persisting through sessions. The case exposes a paradox—routing "dangerous" tasks to weaker models makes errors more likely exactly where precision matters.
More from Fun
- Artist shows off the most beautiful machine they've made — and it sings — techartist_ · 2026-10-16
- OpenAI DevDay hands out Codex 'pet' collector cards, attendees show them off — gabrielchua · 2026-10-16
- Agent named 'Ilya' reportedly scores 1.1pt higher on benchmarks — daniel_mac8 · 2026-10-16
- Viral AI meme: "AI will be safe because nobody is doing the thing they started 2 months ago" — Puzzleheaded-King584 · 2026-10-16
- AI-generated handbag made of vintage keyboard keycaps goes viral, typos included — alysha_lobo · 2026-10-16
- Vibe-coded iPhone Duo sniping bot built with Codex in 17 minutes, 41 tests passing — xiaohu · 2026-10-16