Anthropic Adopts SynthID-Text to Meet EU AI Transparency Mandates

The European Union enacted the "EU AI Act" on August 2, 2026. The law requires artificial intelligence providers to label synthetic content to ensure transparency.

Anthropic and other developers are adopting watermarking practices to comply with these requirements. Available reports do not identify the other companies involved.

1. This move ensures Anthropic complies with the EU AI Act. The act became enforceable on August 2, 2026, and requires providers to label AI-generated content.

2. Anthropic uses "SynthID-Text," a watermarking method developed by Google DeepMind. The system adds no hidden characters to the text. It does not increase the cost of model usage.

3. The company states the watermarking process does not degrade the quality, creativity, or readability of text generated by Claude.

4. Anthropic joined other major artificial intelligence developers to sign a Code of Practice. The agreement establishes uniform watermarking standards across the industry.

1. Business Insider reported that the policy triggered negative reactions. The publication noted that some users on X canceled their Claude subscriptions. Separately, users on Reddit labeled the move a conspiracy.

2. Anthropic stated that its watermarking system performs less effectively for short passages or factual content. The company attributed this to limited word choices available to the model in those instances.

3. Watermarks lack permanence. A complete human rewrite of generated text removes the embedded signal.

4. Watermarks have a minimal effect on code generation. Models possess few arbitrary choices when writing functional code. This limitation reduces the scope of the transparency measure.