New research from Lasso Security found that Google's SynthID-Text watermarking can alter not only word selection but also the tools a large language model invokes and whether it adheres to or disregar…
Lasso Security found that watermarking AI models changes agent behavior, producing differences in tool handling and model refusals, according to The Register. The finding indicates that watermarking t…
Security researchers are warning that the code layer wrapping AI models, known as the harness, is becoming a primary attack surface, with exploits requiring no malicious model behavior. Elad Meged, fo…
Lasso Security reported on August 3 that changing only the agent harness altered outcomes across a 1,000-attack autonomous red-team test. The two harnesses averaged similar objective success rates, ab…