Attackers Steal METR API Key and Consume AI Credits Worth About $600,000
Read original article ↗AI Summary
METR, a non-profit AI research organization, disclosed two security incidents in 2026 involving unauthorized access to its systems. In March, attackers exploited a 'fail-open vulnerability' in a publicly accessible 'vibe-coded app' on a researcher's personal EC2 instance, leading to the exposure of an API key that was used to consume approximately $600,000 worth of AI inference credits. The attackers gained access by scanning certificate transparency logs for high-signal keywords related to LLMs and agents, then directly prompted the agent to reveal the API key. In May, a separate campaign involved systematic probing of METR's infrastructure, including credential stuffing and OAuth token attacks, alongside an inadvertently exposed SQL query endpoint that could have allowed access to unpublished evaluation data, though no evidence indicates data was exfiltrated.
AI-extracted · verify before operational use
No entities or IoCs were extracted from this article.