Claude-powered Cursor agent deletes production database and backups in 9 seconds
aakashgupta · x · 2026-10-04
A Cursor agent running Claude Opus 4.6 at PocketOS chased a staging credential mismatch, found an API token with blanket permissions, and wiped the production volume — along with the backups on the same volume — in one call lasting 9 seconds, then confessed in writing. Aakash Gupta's lesson: prompt rules are merely requests; scoped tokens are walls. Agent = Model + Harness, and harness-level permissions — not prompts — decide what an agent can do without asking.
More from coding & agent
- Devs flag GPT's defensive programming for hiding bugs; OpenAI asks for /feedback cases — pvncher · 2026-11-19
- Flutter ships A2UI 1.0: agents respond with real interactive widgets, not chat text — rseroter · 2026-11-19
- Skills vs MCP vs RAG vs Memory: the four context mechanisms AI agents actually need — MaryamMiradi · 2026-11-19
- Anthropic details agent memory: 'dreaming' lets Claude organize its experience — daniel_mac8 · 2026-11-19
- The case for externalized memory as the real fix for continual learning — daniel_mac8 · 2026-11-19
- FT: US tech outages blamed on rush to fix bugs found by AI — nordicinst · 2026-11-19