Anthropic disclosed that all output from its Claude family of large language models will now carry an invisible watermark, a technical tag designed to comply with the European Union's AI Act. The move, announced on August 13, applies to Claude's standard offerings as well as the code‑focused Claude Code, and it does not alter the meaning, quality, or readability of the text.
"We’re adding marking to Claude’s output to comply with the EU AI Act, and other labs are taking similar steps," the company said in a statement. "It’s hard to identify AI‑generated text, and this gives people better tools for identification. Text from supported Claude models, including output from Claude Code, will carry an invisible watermark, and it doesn’t change the meaning, quality, or readability of Claude’s responses. We also plan to ship a text detection API so users can do more of this themselves."
The EU's new transparency obligations aim to preserve trust in the information ecosystem. European Commission guidelines stress that users should be aware when they are interacting with or consuming AI‑generated content, helping them calibrate trust and avoid deception. Failure to meet these standards could expose companies to penalties of up to €15 million or 3 percent of worldwide annual revenue.
Regulators expect AI firms to pioneer watermarking solutions because the technology evolves quickly and new methods are needed to trace the origin of information. While the EU sets broad criteria—reliability, interoperability, effectiveness, and technical feasibility—it leaves the design of specific marks to the industry.
Anthropic's approach aligns with the EU's call for robust detection tools. By bundling a detection API with the watermark, the company gives developers and publishers a way to verify content provenance on their own platforms. The initiative also signals that leading AI labs are taking the regulatory push seriously, a sentiment echoed across the sector.
If the watermark proves ineffective, the AI Act’s steep fines could quickly become a financial burden. Companies that overlook compliance risk not only monetary penalties but also reputational damage as public scrutiny of AI‑generated misinformation intensifies.
Anthropic's announcement follows a broader industry trend toward greater transparency. While the watermark remains invisible to end users, the accompanying detection API offers a practical path for platforms that wish to label AI‑generated text for readers. The dual strategy—embedding a hidden tag and providing a verification tool—positions Claude as one of the first major models to meet the EU's stringent new standards.
Este artículo fue escrito con la asistencia de IA.
News Factory APP - noticias agénticas para impulsar tu SEO y AEO.