← The Polylog AI Intelligence Brief
Morning Edition · Wednesday, August 12, 2026Published at 2:11 AM EDT · New York
Anthropic Begins Watermarking All Claude Text Worldwide Under an EU Transparency Code
The marking happens at the model level, so it follows output through the interface, the application programming interface and Claude Code, days after researchers published a general method for stripping watermarks from artificial intelligence (AI) images.
Anthropic said it will embed invisible markers in text generated by Claude, applying the change to models launched from 2 August 2026 onward. Because the marking is applied at the model level, it travels with output from the application pro…
Continue the AI Intelligence Brief
Track frontier labs, chips, export controls, model releases, regulation, and AI infrastructure.
- 5 AI intelligence signals a day
- Frontier labs, compute, and chips
- Model releases and AI infrastructure
- Source-grounded analysis with confidence labels
The Global Intelligence Brief stays free.
Part of a tracked trend
Oversight and Evaluation Lag Accelerating AI Capabilities
Over the next 3-6 months, evidence mounts that governance, evaluation, and agent-safety methods are failing to keep pace with capability growth, driving investment in interpretability, agent-manipulation benchmarks, and institutional-reform proposals.
Townsquare
Start a discussion in Townsquare.
More from this edition
- Nvidia Enlists Six Asset Managers to Turn AI Compute Into a Financeable Asset
- OpenAI Puts Its Offensive Cyber Model on Amazon Bedrock Behind a Vetting Program
- Google Reports Its Medical AI Matched Doctors in Simulated Video Consultations
- Nvidia Releases a 31.6-Billion-Parameter Open Model That Activates 3.6 Billion Per Token
- Meta Opens the Weights of Muse Glimmer and Says Its Flagship Coding Model Will Follow
- Anthropic Makes Automatic Approval the Default in Claude Code on 14 August
- OpenAI Extends ChatGPT Advertising to Five More Countries
- Meta's Open Vision Models Cut Beamline Analysis From a Month to Fifteen Minutes at Berkeley Lab
Comments
1Aug 12, 2:00 PM · edited
Model level text watermarking biases token sampling; the researchers attacked post hoc image encoding, which is a different attack surface, and no demonstrated stripping method for the text approach exists.