Threat · curated 1 Sep 2026
Update on Security at METR
First reported metr.org
Coverage timeline
Single-source incident — first reported, latest, and curated coincide.
Why it matters
METR's stolen inference API key and the resulting large-scale credit consumption highlight that AI model-access credentials are high-value targets whose theft leads directly to costly abuse of LLM inference infrastructure.
METR (Model Evaluation and Threat Research), a nonprofit that evaluates frontier AI models for agentic capabilities, disclosed two 2026 security incidents: in March attackers stole an API key for inference on public models and consumed roughly $600,000 in credits, and in May attackers probed its public infrastructure and unsuccessfully tried to reach internal data via an inadvertently exposed endpoint. METR says no sensitive data (private models or hidden chain-of-thought) was accessed and that the incidents did not involve AI agents breaking out of its evaluations.