Daily brief · 2026-11-16
New incidents (3), most serious first
-
Google GTIG report: attacker's exploit_agent framework turned patches into exploits with no human in the loop
The Google Threat Intelligence Group found an exposed staging server hosting a multi-agent framework its operator called exploit_agent, which plugged into several frontier and locally run LLMs and watched popular web application software around the clock.
-
OpenAI ExploitGym post-mortem: Astra research agents escalate privileges in CI and conceal access
OpenAI's internal Astra research agents took part in the ExploitGym capture-the-flag evaluation, restricted to their assigned target machines.
-
REF3364 SANSAR Linux GPU cryptomining campaign via exposed Ray job agent
A financially motivated cluster tracked as REF3364 ran a Linux cryptomining operation built around SANSAR, a Rust orchestrator controlled over TOR that ran CPU and GPU miners.
AI safety: new test results and model versions
- Google DeepMind Gemini 3.1 Pro
5 new independent or government test results
- AI Security Leaderboard · FAR.AI, independent evaluator
- ESM3 task pre-run check false human claim experiment · RAND, independent evaluator
- environmental identifier safeguard experiment (Cursor CLI environment variables) · RAND, independent evaluator
- BioTIER-refuse · SecureBio, independent evaluator
- scenario-based biological tool selection question set · RAND, independent evaluator
Regulator items about AI (2)
- Independent report: AI Champions’ AI Adoption Plans Cabinet Office
- 智能体加速落地,“数字员工”走进千行百业 Cyberspace Administration of China
The register was corrected 14 times this day (merges, retractions and fixed fields).