Anthropic is among 234 AI providers that have adopted the EU AI Act's voluntary transparency code, embedding watermarking in AI outputs to meet new regulations. The move highlights the industry's shift towards multi-layered transparency approaches.
TL;DR
- Anthropic and 233 other AI providers have adopted the EU AI Act's voluntary transparency code, embedding watermarking in AI outputs.
- The EU AI Act's transparency obligations require AI providers to implement machine-readable detectability or watermarking for AI outputs by December 2, 2026.
- The industry is moving towards a multi-layered approach to transparency, combining metadata, watermarking, and content provenance.
What happened
The EU AI Act's transparency obligations, which came into force on August 2, 2026, require AI providers to implement machine-readable detectability or watermarking for AI outputs by December 2, 2026.
Anthropic, the developer of the Claude AI system, has published its approach to meeting these obligations. All Claude models launched on or after August 2, 2026, will embed invisible watermarking in generated text and digitally signed provenance metadata in generated images.
Google DeepMind has been an early adopter of watermarking, embedding invisible watermarks into images, audio, text, and videos deployed across Google products. Microsoft offers users the option to add visible watermarks on generated pictures or audio, and OpenAI uses C2PA metadata for images and has partnered with Google DeepMind to embed SynthID watermarking in its AI-generated image outputs.
Why it matters
The adoption of the EU AI Act's transparency code by Anthropic and other AI providers highlights the industry's shift towards multi-layered transparency approaches, combining metadata, watermarking, and content provenance.
The Italian Data Protection Authority's warning against a broadcaster for airing satirical, AI-generated deepfakes without adequate marking illustrates the importance of complying with the deployer transparency obligations.
The UK Jurisdiction Taskforce's final Legal Statement on Liability for AI Harms under the private law of England and Wales provides certainty and predictability for businesses using AI, emphasizing the need for proactive risk management, robust contractual protections, professional oversight, and clear statements and disclaimers when using AI to communicate with customers, partners, and the public.
Key facts
- Anthropic, Google, Meta, Microsoft, Mistral, and OpenAI are among the 234 organizations that have signed up to the EU AI Act's voluntary transparency code.
- Anthropic's Claude models launched on or after August 2, 2026, will embed invisible watermarking in generated text and digitally signed provenance metadata in generated images.
- Google DeepMind has been embedding invisible watermarks into images, audio, text, and videos deployed across Google products since 2023.
- Microsoft offers users the option to add visible watermarks on generated pictures or audio.
- OpenAI uses C2PA metadata for images and has partnered with Google DeepMind to embed SynthID watermarking in its AI-generated image outputs.
- The EU AI Act's transparency obligations require AI providers to implement machine-readable detectability or watermarking for AI outputs by December 2, 2026.
- The Italian Data Protection Authority issued a warning against a broadcaster for airing satirical, AI-generated deepfakes without adequate marking, highlighting the importance of complying with the deployer transparency obligations.
- The UK Jurisdiction Taskforce's final Legal Statement on Liability for AI Harms under the private law of England and Wales emphasizes the need for proactive risk management, robust contractual protections, professional oversight, and clear statements and disclaimers when using AI to communicate with customers, partners, and the public.
Context
The EU AI Act's transparency obligations are part of a broader effort to regulate AI and ensure that AI systems are transparent, explainable, and accountable.
The Italian Data Protection Authority's warning against a broadcaster for airing satirical, AI-generated deepfakes without adequate marking highlights the potential risks and harms associated with AI-generated content.
The UK Jurisdiction Taskforce's final Legal Statement on Liability for AI Harms under the private law of England and Wales provides guidance for businesses using AI, emphasizing the need for proactive risk management, robust contractual protections, professional oversight, and clear statements and disclaimers when using AI to communicate with customers, partners, and the public.
