Hiding Capabilities in AI Models
Interesting research: " AI Models Can Provably Hide Arbitrary Capabilities " Abstract: Capability evaluations aim to surface dangerous model behaviors before deployment, but their reliability depends
Search Intelligence
Backporting security fixes to stable releases
The software you are running is almost certainly riddled with known, unfixed security bugs. These bugs can be trivially and quickly exploited by modern AI agents. This is not merely theoretical; I saw
Experts temper emergency-level alarms over critical Brotli defect
The defect in a widely distributed open-source library provoked an emergency response, yet exploitation hasn’t been observed in the wild. The post Experts temper emergency-level alarms over critical B
Bias in Qwen3-Coder Model
Interesting research : TL;DR: Chinese AI models now sit inside Western products, and some carry geopolitical triggers: tell the model who the code is for, and the security of what it writes changes. F