AI model watermarking changes agent behavior

According to Lasso Security, AI model watermarking changes how AI agents handle tools and safety refusals. The altered behavior isn't necessarily worse but can be, particularly under adversarial ...

Published: 2026-09-17