Facebook
Britain's News Portal
Around The Clock
BREAKING
Loading latest headlines…

AI model watermarking changes agent behavior, Lasso Security finds

Lasso Security reports differences in tool handling and model refusals when AI models are watermarked.

  • Lasso Security observed differences in tool handling and model refusals linked to AI model watermarking.
  • The findings concern changes in agent behavior.

Watermarking AI models can change how agents behave, according to Lasso Security. The security firm reports seeing differences in tool handling and in model refusals.

The observation concerns agent behavior rather than model outputs alone, with the reported changes affecting how agents handle tools and when models decline requests.

Lasso Security's findings were reported on 17 September 2026.

Related Articles

Get the news that matters.

Join thousands of readers getting the best of British news straight to their inbox.