Intel / post

How a Vibe-Coded Dashboard's Auth Flaw Let an Attacker Drain $600K in Credits

Source: AlphaSignal
Date: Published Sep 22, 2026

Key takeaways · AI-distilled

Per AlphaSignal's account, METR's vibe-coded dashboard exposed an through an auth flaw; an attacker prompted it to reveal its API key and used about $600K in credits over three weeks. The provider covered them and METR reported no loss.
An HTTP 401 does not prove nothing happened: if the backend queues work before authentication, a worker can still execute it. The post says to check the response, the queue, and the task database together.
Trace which credentials the worker can read. Holding the provider key in a separate service lets that service enforce access and usage limits instead of exposing the key to code-execution workers.
The post's pre-deploy checks: confirm rejected requests create no jobs, test access across different user accounts, keep provider keys out of code-execution workers, restrict worker permissions, and test credential revocation and recovery.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.

Why it matters

Lays out a replicable failure pattern in agent dashboards, unauthenticated requests still queuing jobs a worker executes, and gives concrete checks to catch it before shipping.

Read the source x.com