Anthropic will begin to watermark any materials – text and images – produced by its AI to comply with new AI rules from the EU.

The AI firm will leave invisible markers to show that text and images have been AI generated as new EU AI Act rules come into effect.

The watermarks will be invisible to the naked human eye, but can be discernible by automated systems in order to know what material may have been generated by AI.

The updated rule comes amid the rise in rampant low-quality AI images and videos, colloquially but not affectionately known as ‘AI slop’, across the internet. Further, the spread of misinformation and disinformation has caused a scourge on a society attempting to evade a collective AI-induced psychosis as people struggle to tell what is real and what is AI-generated.

Now, Claude has long introduced a watermark to AI-generated images – while effective, these largely disappear if someone takes a screenshot and goes on to share the reproduced image.

However, Claude text will now carry an invisible watermark, which is prompting concern. Some users of AI fear that a watermark may mark their work as AI-generated even if they go to Claude simply for edits, despite their words and ideas being original.

Anthropic says that this remains a major limitation of this system – “A detected mark provides a signal that content was processed by Claude, but is not fully conclusive” the AI firm said.

This means that the watermark does not say that the image or text was completely AI-generated, but equally a lack of a mark does not mean materials were not processed through AI.

Recommended

• Meta Announces Plans For World’s Largest Sub-Sea Cable
• Workers Spend 3 Hours A Week Dealing with ‘AI Workslop’
• 65% of Employees Are Excited to Use AI at Work
• Seven In Ten UK Employees Using Unapproved AI Tools, Microsoft Finds
• Report: AI Adoption Elevates Human-centric Skills
• Will AI Enable Workers to Secretly Juggle Multiple Jobs?

It may have been heavily edited post-processing, or be too short to have a strong enough AI signal for watermarking, Anthropic said.

While Anthropic did not disclose how it would be rolling out this watermark over text, it said that the inclusion would not “change the meaning, quality, or readability of Claude’s response.”

All models that were released after August 2nd will generate content with a watermark, the firm said, though it is working on supporting older models with the remit.