AI watermarking could make LLM guardrail adherence unpredictable and that could be a big problem for the EU AI Act

Date:
Fri, 18 Sep 2026 12:09:44 +0000

Description:
Researchers have revealed how AI watermarking could actually change how an
LLM behaves, making it less secure.

FULL STORY ======================================================================Copy link Facebook X Whatsapp Reddit Pinterest Flipboard Threads Email Share this article 0 Join the conversation Follow us Add us as a preferred source on Google Newsletter Subscribe to our newsletter AI watermarking aims to prove the authenticity of any text New study finds it also changes LLM behavior
and in a bad way EU AI Act could mean more models have watermarks despite
side effects New Lasso research has revealed that AI watermarking could actually unintentionally change how LLMs behave following the testing of Google DeepMind's SynthID-Text.

The company's researchers found that SynthID-Text can change whether models refuse harmful requests, their susceptibility to prompt injection, which
tools AI agent choose and more. However, at its core, SynthID-Text and other similar watermarking is only designed to hide a machine-readable indicator as to whether text was AI-generated or human-written. Latest Videos From TechRadar Watch full video here: Researchers find that AI watermarking can unintentionally change AI behavior The "watermarking procedure can therefore affect both what the model says and what an agent does," Lasso concludes, referring to the side effect as "sampling drift."

One of the biggest concerns highlighted by the paper is that, even without an attack, watermarking changed some of the models' refusal decisions, making them more willing to answer potentially harmful prompts. Combined with prompt injection, Lasso found the consequences more amplified. You may like Claude will now hide an invisible watermark inside ordinary words Shadow AI is a security problem, but the EU AI Act makes it a legal one Irregular AI lab spots agents switching models without humans instruction in agentic self-modification phenomenon

Despite the unintended consequences, Anthropic recently announced that future generations of Claude would use AI watermarking similar to Google DeepMind's, stressing that one of the key drivers was to adhere to the EU AI Act. With that in mind, AI watermarking is set to become far more mainstream across other model providers, making these mishaps far more common and leading to further security concerns.

Ultimately, Lasso urges developers to rerun benchmarks, safety evaluations
and other tests to check for any unintended consequences, rather than just applying it blindly to existing configurations. Are you a pro? Subscribe to our newsletter Sign up to the TechRadar Pro newsletter to get all the top news, opinion, features and guidance your business needs to succeed! Contact me with news and offers from other Future brands Receive email from us on behalf of our trusted partners or sponsors By submitting your information you agree to the Terms & Conditions and Privacy Policy and are aged 16 or over.

"These findings make reassessment important whenever watermarking is introduced or its configuration or key changes," Lasso concludes, stressing that the work shouldn't be taken as an argument against AI watermarking for provenance.

Additionally, the research presents a new angle on AI watermarking, because until now researchers have largely focused on whether watermarks can be both applied and detected effectively. Few have uncovered such security-focused consequences as this one. Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.



======================================================================
Link to news story: https://www.techradar.com/pro/ai-watermarking-could-make-llm-guardrail-adheren ce-unpredictable-and-that-could-be-a-big-problem-for-the-eu-ai-act


--- Mystic BBS v1.12 A49 (Linux/64)
* Origin: tqwNet Technology News (1337:1/100)