Cyber Risks · The Hacker News

Attackers Steal METR API Key and Consume AI Credits Worth About $600,000

Summary: METR disclosed two security incidents: in March 2026 attackers stole an API key for public-model inference and consumed credits that would have totaled about $600,000 had the model provider not supplied them free, and in May 2026 attackers systematically probed its public infrastructure, including an unsuccessful attempt to reach internal data through an inadvertently exposed endpoint. The first attack traced to a researcher's vibe-coded agent dashboard on a public EC2 instance whose fail-open vulnerability silently disabled Google authentication for several days; the attacker prompted an agent to reveal its model provider API key, added an SSH key for persistence, and used the credentials for three weeks undetected because METR's large evaluations consume high token volumes and no spend caps were set.

Bottom line: The theft went unnoticed for three weeks because heavy token consumption was normal and no spend caps existed - a $600,000 bill absorbed only because the model provider waived it. A fail-open authentication bug on one researcher's side project was enough to expose credentials to an attacker scanning certificate transparency lists for agent-related sites.

Read the original at The Hacker News ↗

Topics
Data Breaches
Related