gg2
Claude AI news
Archived — this story has rotated out of today’s deck. It is kept here in full.
The gist
Anthropic began watermarking all Claude text outputs on August 2, 2026. Users are canceling subscriptions over fears of AI-labeling.
Background
Anthropic, the maker of Claude, announced that all Claude models would embed an invisible watermark in generated text to comply with the EU AI Act's transparency requirements. The watermark is designed to be imperceptible but detectable, using a technique called SynthID-Text. This move has sparked debate among users and developers, with some concerned about the implications for authorship and privacy.
How it unfolded
- Aug 2, 2026Anthropic begins embedding invisible watermarks in text generated by Claude models launched in the EU, with plans to apply worldwide.
- Aug 13, 2026Anthropic publishes a support page explaining the watermarking, but initially provides few technical details.
- Aug 14, 2026Some Claude users cancel subscriptions, citing concerns about the watermark affecting professional and academic work.
- Aug 15, 2026Anthropic shares more details, revealing the use of Google DeepMind's SynthID-Text approach and plans for a detection API.
- Aug 16, 2026Claude experiences a major outage affecting Claude.ai, Claude Code, and Claude Cowork, with authentication issues and degraded performance.
Who’s saying what
- Support
- Anthropic argues that watermarking provides transparency and helps people understand the provenance of AI-generated content.
- Caution
- Some users worry that the watermark could lead to false accusations of AI use in academic or professional settings, or trigger contract penalties.
- Critics
- John Gruber of Daring Fireball calls the watermarking a 'perversion of writing' because it alters word choices and can degrade text quality.
Still unverified
The exact cause of the August 16 outage is still under investigation by Anthropic.