Anthropic says new Claude models will include AI-content markers, while files will use the C2PA standard to improve transparency and traceability.
San Francisco: Anthropic is introducing new measures to identify content generated or edited by its Claude AI models, as the European Union’s new AI transparency requirements take effect.
Under the EU AI Act, providers of certain AI systems are required to ensure that AI-generated or manipulated content can be identified in a machine-readable manner. Anthropic has outlined how it plans to implement these requirements through its updated Claude support documentation.
How Will Claude’s Watermarking Work ?
Anthropic says new Claude models launched from August 2, 2026, will include technology designed to mark AI-generated content.
The system will apply not only to text but also to generated files. For files, Anthropic says it is using C2PA, an open technical standard designed to provide provenance information about digital content.
The company also plans to extend the capability to older Claude models where supported.
For text, the marking is designed to remain associated with content when it is copied and pasted into other applications. However, Anthropic acknowledges that the signal may not survive certain transformations, such as changing the formatting, taking a screenshot or substantially rewriting the content.
A Watermark Does Not Automatically Prove Claude Wrote Everything
Anthropic has also highlighted an important limitation.
The presence of an AI-content marker does not necessarily mean that Claude originally wrote every word in the content. Users can submit human-written material to Claude for purposes such as editing, summarisation or translation.
Therefore, a marker may indicate that an AI system was involved in processing the content, rather than proving that the entire piece was generated from scratch by Claude.
Applies Across Claude Products
Anthropic’s AI-content marking approach is intended to cover its broader Claude ecosystem, including Claude API, Claude App, Claude Code and Claude Cowork, subject to the capabilities of individual products and models.
The move reflects a broader industry effort to make AI-generated and AI-assisted content easier to identify.
EU AI Act Drives Greater Transparency
The EU’s AI regulatory framework is placing increasing emphasis on transparency, content provenance and disclosure of AI involvement.
The objective is to make it easier for users, platforms and other systems to distinguish AI-generated or AI-manipulated material from conventional human-created content.
Companies that fail to comply with applicable provisions of the EU AI Act can face significant regulatory penalties, although the specific obligations and enforcement mechanisms vary depending on the type of AI system and the provision involved.
Other AI Companies Also Exploring Content Identification
Anthropic is not alone in responding to growing regulatory and industry pressure around AI-content transparency.
Companies and platforms including Google, Meta, Microsoft, OpenAI, Synthesia and Black Forest Labs have been working on provenance, watermarking or AI-content identification technologies.
AI music platform Suno has also explored marking AI-generated tracks, while newsletter platform Substack has partnered with Pangram on AI-content detection.
What It Means for Claude Users
The development marks a shift toward AI systems where content provenance becomes part of the generation process itself.
For users, the key distinction is that AI marking is intended to provide a signal that AI was involved—it is not necessarily an infallible AI detector or proof of complete AI authorship.
As regulatory requirements expand, such provenance technologies are likely to become increasingly important for publishers, businesses, educational institutions and online platforms dealing with AI-generated content.
