AI-content transparency is quietly becoming a real, ongoing compliance question rather than a one-off headline. In August, Anthropic began watermarking text generated by Claude, and around the same time Meta expanded its AI content labels to cover images and videos edited with third-party AI tools like Canva, not just fully AI-generated posts. For brands that use AI anywhere in their content pipeline, disclosure is becoming routine rather than optional.
Claude’s Text Watermarking
Anthropic’s watermarking, built on Google DeepMind’s SynthID-Text method, works by subtly influencing which words Claude selects when multiple choices are equally valid. The pattern is invisible to a reader but verifiable with a detection key, and Anthropic offers that detection API to regulators, researchers, and fact-checkers. The move came after Anthropic signed the EU’s Code of Practice on Transparency of AI-Generated Content, part of roughly 190 signatories now working toward EU AI Act compliance. Internal testing found no impact on content quality, creativity, or readability.
Meta’s Expanded Labels
Meta has been broadening what counts as “AI-generated” for labeling purposes, moving beyond content made entirely by AI to include material edited or enhanced with AI-powered tools, including third-party apps like Canva. That is a meaningfully wider net than most brands are used to thinking about when they consider whether a piece of content needs a disclosure label.
What This Means for Your Content Pipeline
If your team uses AI anywhere in the process, from a first draft to a touched-up product photo, it is worth mapping out where that content ends up and whether the platform’s disclosure rules apply. This connects directly to the broader move toward visible imperfection in design we covered earlier this year: as labeling and detection both get more sophisticated, brands leaning into authentic, clearly-human touches have less to worry about than those trying to pass AI-assisted work off as fully organic.
Frequently Asked Questions
Does Claude’s watermark change how the text reads?
No. Anthropic’s internal testing found no measurable impact on content quality, creativity, or readability from the watermarking process.
Does Meta’s expanded AI label apply to content just edited with AI, not fully generated by it?
Yes. Meta’s labeling now extends to content edited or enhanced with AI-powered tools, including third-party apps, not only content created entirely by AI from scratch.
Why are companies doing this now?
It is largely driven by regulation, particularly the EU AI Act and the EU’s Code of Practice on Transparency of AI-Generated Content, which a growing number of AI providers and platforms have committed to.
Need a content and branding strategy that holds up under new AI-disclosure rules? Visuapex Creatives builds branding and graphic design that stays authentic by design. Get a free consultation.
Source: Anthropic.