Hacker NewsMETR: SWE-bench Passing PRs Would Not Be Merged — Benchmark Validity CrisisMETR·high signalMETR finds many SWE-bench passing PRs wouldn't be merged by humans. Fundamentally challenges the primary AI coding benchmark.SourceSource pageMETR↳ Follow the threadShared entity / Stack layerCoding-agent benchmark reality check: Codex CLI (GPT-5.5) tops Terminal-Bench 2.1 at 83.4%, Devin's 'real' SWE-bench nearer 9–10%CodeSOTAShared entity / Stack layerMETR: GPT-5.6 Sol Sets Record for Reward-Hacking, Making Its Benchmarks 'Unreliable'Tech TimesStack layer / ContrastGodot Bans Almost All 'Vibe-Coded' and Autonomous-Agent Contributions to Protect Volunteer MaintainersThe RegisterPolicy dependency / Stack layerIAPS warns on Kimi Claw: security and governance risks of Chinese-hosted 'always-on' AI agentsIAPSStack layer / Update threadLatent Space: Modal CTO on Why AI Infrastructure Must Evolve for 'Agent Experience'Latent SpaceStack layer / Update threadAmjad Masad Details Replit's Nightly Self-Improving Agent That Reads Every Trace and Ships Its Own A/B TestsSaaStrStack layer / ContrastFortune confirms Anthropic has overtaken OpenAI on revenueFortuneStack layer / Threat patternArmadin details full sandbox escape in Anthropic's Claude Cowork; Anthropic disputes, silently patchesSiliconANGLE