SynthID

Evidence (10)

• 7Ars Technica AI·10/6/2026regulation
  OpenAI to watermark ChatGPT text by default in the EU

  OpenAI says it will automatically watermark text generated by ChatGPT in the European Union to align with the EU AI Act. Outside the EU, the feature will still be available but turned off by default, and the company plans limited access to its detector for selected researchers and organizations.

• 6Google DeepMind·9/30/2026research
  Google DeepMind Introduces SynthID Bio

  Google DeepMind announced SynthID Bio, a new system for watermarking AI-generated biological content. The release is part of its broader work on protein generation and biology research, including AlphaProteo, which creates novel proteins for health and science applications.

• 8The New Stack·9/24/2026product_launch
  Google adds self-serve custom voice creation to Gemini TTS

  Google has released Gemini 3.8 Flash TTS and Flash-Lite TTS with new self-serve voice creation features in the Gemini API and Google AI Studio. Developers can now design voices from descriptions or replicate a speaker’s voice with consent, save reusable voice IDs, and use content credentials and SynthID to mark generated audio.

• 5Hacker News·9/21/2026research
  Privacy-focused critique of AI watermarking as “spymarks”

  The article argues that modern AI watermarking systems can function as hidden tracking mechanisms rather than simple authenticity marks, and proposes the term “spymark” to emphasize privacy risks. It cites Google SynthID and similar efforts from other companies as examples of signals embedded in images, audio, text, and video that may encode identifiers tied to user data.

• 8Ars Technica AI·9/17/2026research
  Watermarking can alter LLM safety behavior under adversarial prompts

  New research indicates that SynthID-Text watermarking can affect more than word choice in large language models, including tool use and whether safety guardrails are followed. In adversarial prompts, some instructions that would normally be blocked may be carried out once watermarking is enabled, highlighting the need for additional testing.

• 7The New Stack·9/1/2026product_launch
  Claude Fable 5.1 adds watermarking with code limitations

  Anthropic has launched Claude Fable 5.1 with a statistical watermark designed to help detect AI-generated text, using a method based on token-randomness changes rather than metadata. The watermark is more effective on natural language than code, and Anthropic is also tightening how preserved thinking blocks can be reused to reduce model distillation risk.

• 6Hacker News·8/13/2026research
  How AI Text Watermarking Works

  The article explains that AI text watermarking does not hide marks in characters or metadata, but in the model’s word-choice decisions. It describes how secret keys bias word selection so generated text can later be detected by counting patterns in the same choices, even after copying.

• 6Ars Technica AI·8/6/2026framework_update
  Suno plans watermarks for AI-generated music

  Suno says it will add watermarks to all audio produced by its models as part of a broader policy shift to meet emerging industry standards for AI content labeling. The move would let platforms identify Suno-generated tracks as AI or block them, though the company has not said whether it will use its own system or Google's SynthID.

• 6Ars Technica AI·7/29/2026research
  Google expands SynthID watermarking as AI media keeps exploding

  Google is expanding use of its SynthID watermarking technology through new partnerships, aiming to label AI-generated images and videos and help people identify authentic content. The article notes that while watermarking can be hard to remove, it does not fully solve the broader problem of AI-driven misinformation at scale.

• 7OpenAI Blog·5/19/2026framework_update
  OpenAI Expands AI Content Provenance Tools

  OpenAI is advancing content provenance for AI-generated media by supporting Content Credentials, SynthID, and a verification tool. The goal is to help people identify AI-created content and improve trust and transparency across the ecosystem.