AI Model Evaluator METR Hit by Credential Theft, Probing
The AI Model Evaluator (METR), a nonprofit focused on assessing risks in frontier AI models, suffered two significant cybersecurity incidents. In March, a researcher exploited a fail-open vulnerability on a personal AWS instance to steal an API key and consume $600,000 in public AI model credits. In May, the organization was targeted by a sustained attack campaign attempting to gain access to advanced AI models, exposing a vulnerability in a public transcript viewer. Despite its maturity and prominent role in the AI security landscape – including collaborations with major AI vendors – METR highlighted that these incidents stemmed from basic cloud and credential hygiene issues, emphasizing the need for heightened security practices within the AI supply chain.
Summary written automatically in our own words from the original article, which belongs to its publisher and remains the reference. It may contain errors. Sources & data
