Evaluation & Safety
How agents fail and who catches it: benchmarks, red team results, incident reports, and the scores that decide what ships.
UK Government Testers Got Claude Opus 5 to Breach a Test Enterprise Network in 8 of 10 Attempts
Evaluation & Safety · 5 minFirst Joint UK-US Assessment Finds Kimi K3 Far Behind US Models on Offensive Cyber
Evaluation & Safety · 5 minOpenAI Paused Its Erdos Model After It Escaped the Sandbox
Evaluation & Safety · 4 minFour AI Agent Attacks, One Root Cause: The Infrastructure
Evaluation & Safety · 4 minOpen Models Are Four to Seven Months Behind the Cyber Frontier, and Nearly Free
Evaluation & Safety · 4 minMedicare's AI Prior Authorization Pilot Is Under Fire, and It Runs Through 2031
Evaluation & Safety · 4 minA Record Patch Tuesday Fixes a 9.6 Copilot Bug as Microsoft Credits AI for the Flood
Evaluation & Safety · 4 minAn Unpatched Claude for Chrome Flaw Lets Any Extension Puppet the Agent
Evaluation & Safety · 4 minGhostcommit Hides Agent Attacks Inside a Pull Request's Images
Evaluation & Safety · 4 min99.9 Percent of Fixable AI Vulnerabilities Are Sitting Unpatched in Production
Evaluation & Safety · 4 minAttackers Hijacked an AI Gateway With a Path to Amazon Bedrock
Evaluation & Safety · 3 minHalluSquatting: Attackers Register the Packages Your Coding Agent Dreams Up
Evaluation & Safety · 4 minFirst Agent Platform on CISA's Must-Patch List Hits Its Federal Deadline
Evaluation & Safety · 3 minGPT-5.6 Shipped a Week After Its Evaluators Said It Games the Test
Evaluation & Safety · 4 minOne word beat the guardrails: GitLost made GitHub's agent leak private repos
Evaluation & Safety · 3 minNine AI companies got a safety report card. The best grade was a C+
Evaluation & Safety · 3 minA safety benchmark just failed every agent it tested
Evaluation & Safety · 3 minMore safety coverage is on the way. Start with What is GaaS? and the glossary of the agent economy.