Brad Keoun of Distro Media argues for ditching Claude over AI text watermarking, in favor of open-source alternatives

Keoun writes that Anthropic announced Aug. 14 it would watermark Claude's generated text using a technique similar to Google's SynthID, and that OpenAI disclosed a similar move earlier this month, both citing EU transparency rules effective in early August. He argues watermarks could expose journalists to bias claims and substitute words he did not choose, though Anthropic says the method has no practical impact on output quality. He suggests locally run open-source models, which can be inspected, as an alternative.

✨ AI summary of the linked source — not editorial content.

Explore the topic

• Claude
• AI
• Media