Anthropic explains how Claude’s invisible text watermarks will work
Anthropic will use Google DeepMind’s SynthID-Text to embed invisible watermarks in Claude-generated text, aligning with EU AI Act transparency requirements.
- Anthropic will use Google DeepMind’s SynthID-Text to embed invisible watermarks in Claude-generated text.
- The watermarking system alters word probabilities to create detectable patterns without affecting readability.
- This aligns with the EU AI Act’s transparency requirements for synthetic content.
- Anthropic is also adding C2PA support for images processed by Claude to enhance traceability.
Anthropic has outlined its plan to embed invisible watermarks in text generated by its Claude AI model, using a version of Google DeepMind’s open-source SynthID-Text technology. The watermarking system works by subtly altering the probabilities of word choices in the output, creating detectable patterns that can be identified without disrupting readability. This approach is designed to help users and regulators distinguish between human-written and AI-generated content, a key requirement under the European Union’s AI Act.
The move follows Anthropic’s broader push to meet compliance standards ahead of the AI Act’s enforcement. Alongside text watermarking, the company is also integrating C2PA support for images processed by Claude, further enhancing traceability of synthetic media. These measures reflect a growing industry trend toward embedding transparency tools directly into AI systems, rather than relying on post-hoc detection methods.
The SynthID-Text watermarking method was first introduced by Google DeepMind as part of its SynthID suite, which aims to make AI-generated content more identifiable. By adopting this technology, Anthropic is leveraging an existing, peer-reviewed approach rather than developing a proprietary solution from scratch. This could set a precedent for other AI developers seeking to balance innovation with regulatory compliance.
Provides a practical, open-source method for embedding watermarks in AI-generated text.
Helps companies comply with EU regulations while maintaining trust in AI-generated content.
Signals Anthropic’s commitment to regulatory compliance, potentially influencing market confidence.
Increases transparency in AI-generated text, addressing concerns about misinformation.
- SynthID-Text
- An open-source watermarking technology by Google DeepMind that embeds detectable patterns in AI-generated text by altering word probabilities.
- C2PA
- Coalition for Content Provenance and Authenticity, a standard for tracking the origin and authenticity of digital content.
- EU AI Act
- A regulatory framework by the European Union to govern the development and use of artificial intelligence systems.
SecuritySIP: Five Immediate Software Supply Chain Controls
SecurityThe Defender’s Window
You thought you were talking to AI. Police may see a digital diary - calcalistech.com
SecurityRogue AI aren’t science fiction anymore
SecurityWildfire smoke now bigger prenatal threat than human sources of air pollution
George Fox Launches Free AI Certificate Courses - georgefox.edu
George Fox University introduced three free AI certificate courses available online, aiming to make AI education accessible to everyone.
Pioneering AI and Digital Transformation Scholar Ravi Bapna Named Dean of Santa Clara University’s Leavey School of Business - Santa Clara University
Santa Clara University names Ravi Bapna, a leading AI and digital transformation scholar, as the new dean of its Leavey School of Business.
When AI Regulation Becomes a Systems Bottleneck - cacm.acm.org
A new ACM report argues that increasingly complex AI regulations are creating bottlenecks in system development and deployment.
BusinessAmazon, which started off selling books, is destroying rare texts to train AI
Amazon has allegedly destroyed rare and out-of-print books to use their text for training AI models, raising ethical and legal concerns.
AI wealth is fueling housing market frenzy in San Francisco: 'A million dollars over asking' - ABC7 Bay Area
Tech workers in San Francisco are paying over $1 million above asking prices for homes, driven by AI industry salaries.
Build OpenClaw agents that transact with Amazon Bedrock AgentCore payments - Amazon Web Services (AWS)
AWS integrates Amazon Bedrock AgentCore payments with OpenClaw agents, allowing autonomous AI agents to execute financial transactions directly.