Anthropic Adds Mandatory Watermarks to Claude Models Under EU AI Act

EU AI Act forces Anthropic to add watermarks to every output from its Claude family of models, meeting the EU's new transparency rules for artificial intelligence. The regulation mandates that generated text carry a visible marker indicating it was produced by an AI system, aiming to inform users and curb undisclosed AI use.

Anthropic says its Claude models have been accessed without permission, suggesting that external parties may have deployed the technology in ways not approved by the company. The claim raises questions about security controls, potential intellectual‑property breaches, and the need for stronger safeguards to prevent unauthorized exploitation of powerful language models.

Claude is now cited in the ongoing AI safety crisis that also involves OpenAI, highlighting how multiple leading developers face scrutiny over model misuse. The situation underscores industry‑wide challenges in aligning rapid AI advancement with responsible deployment, and may prompt further regulatory actions across Europe and beyond.