AI coding agent deletes production database in 9 seconds; Keycard calls for task-based agent auth
matt_slotnick · x · 2026-10-15
Ian Livingstone of Keycard AI argues that authentication and authorization must be redesigned for autonomous agents. He cites a real incident: an AI coding agent fixing a staging credential mismatch found an unrelated Railway API token and used it to delete a production database volume—backups included—in nine seconds. The token worked as designed, but its scope far exceeded the task, and nothing on Railway's side could tell the deletion was unrelated.
Key points from the post:
- Nearly all auth systems assume whoever holds a credential can do anything it allows—an assumption that breaks down when the holder is an agent deciding actions on the fly, often unsupervised for hours or days.
- Agents need broad capability, so safety requires a straight line from intent to action to access: every action gets only the access its task requires, only for as long as needed.
- Like the web and cloud before it, the agent shift demands a new identity and access model; he illustrates it with a 24/7 incident-response agent scenario.
More from coding & agent
- E2B sandbox usage doubled to 2B in 5 months, 61% of traffic now from agents — FinanceYF5 · 2026-10-15
- Context Inertia: agents won't let go of dead-end hypotheses — the fix is throwing out context — dbreunig · 2026-10-15
- Anthropic Dropped Git After Merges Hit 30,000 a Day — shashib · 2026-10-15
- One sentence changed the whole strategy — and every downstream agent updated itself — FellMentKE · 2026-10-15
- Opus Suggests Letting Sol Invent a Geometry Dash Solver to Test Coverage — mgostIH · 2026-10-15
- Experienced Engineers Agree: AI Coding Raises Demand for Engineers, Not Lowers It — bendee983 · 2026-10-15