Anthropic adds machine-readable marks to Claude output for EU users as the AI Act takes effect. California's own transparency law looms for 2025, forcing financial firms to embed detection controls.
Anthropic began embedding machine-readable watermarks into output from its Claude model, complying with the European Union's AI Act transparency rules that took effect Aug. 2. The company outlined the change on its support page Tuesday.
The labels take the form of metadata and digital signatures embedded in generated text and files. Anthropic said it is adding these marks to Claude models launched in the EU on or after the effective date, covering output accessed through its cloud partners at Amazon, Google and Microsoft.
"As AI-generated content becomes commonplace, greater transparency and signals about where content comes from can give people useful context about the information they consume," Anthropic wrote. The company said the move supports its legal obligations under the EU's AI Act.
The EU rules are the first major force to require technical detection signals, not just disclosure labels. California is preparing its own AI Transparency Act for next year. PYMNTS reported last week that the California law, though narrower in scope, pushes the same direction: organizations must embed technical controls into AI content creation and distribution, not simply tell users when AI was used.
For regulated financial institutions using generative AI in customer service, fraud detection, document preparation and internal operations, the shift adds a new governance layer. The PYMNTS report said banks already operate under model risk management, cybersecurity and operational resilience expectations, and AI transparency obligations increasingly intersect with each of those existing frameworks.
The practical effect is that AI vendors like Anthropic must build detection infrastructure at the model level. That infrastructure then flows downstream to enterprise customers, who must incorporate it into their compliance systems. The watermarking applies to Claude output across web, API and cloud channels, reducing the burden on each individual firm to develop its own detection method.
Anthropic said the marking is active for all Claude models launched in the EU after Aug. 2, with support for "digitally signed provenance metadata where available." The company did not give a timeline for extending the watermarks to models outside the EU, though the California law's 2025 effective date is likely to push broader adoption.
Drafted by a large language model from the source reporting linked above, then screened by automated publishing checks. It is not read by a journalist before publication. Some articles cite our Alpha Score. Verify prices and figures against the original source. Educational coverage, not personalized advice.