← Back

Wire · opportunities

AI Model Evaluator METR Hit by Credential Theft, Probing

Published

2 September 2026

Topic

opportunities

Sectors

AI Frontier Models

Geography

United States

Source

Read at darkreading.com

Verified

Fusion42 · 2 September 2026 · Fusion42 review

METR, a security nonprofit evaluating risks in frontier AI models, suffered credential theft leading to $600,000 in API credit misuse, and faced probing for unpublished evaluation data via exposed endpoints. The incidents prompted METR to enhance security controls including credential rotation, monitoring, and shutting down legacy infrastructure.

This Wire brief sits within Fusion42's coverage of AI Frontier Models.

◆ The Wire takeaway

Your AI platform's security must tighten immediately after METR's API key theft showed how attackers can exploit neglected endpoints to cost hundreds of thousands. Rotate credentials aggressively and audit all external access, or risk a costly breach that will hit trust and compliance.

Coverage

1 source · 2 Sep 2026

Related on Wire

Topics

AI Frontier Modelsai-securitycredential-theftapi-abusefrontier-aisecurity-incident