Anthropic’s EU AI Act Compliance Sparks Backlash Over Claude Watermarks
As Anthropic implements invisible watermarks to meet European regulatory requirements, a divide emerges between those defending transparency and users arguing the measure penalises average AI adopters.
Anthropic has introduced invisible watermarks into the editorial text outputs of its Claude AI model to ensure compliance with the European Union AI Act’s Transparency Code. This regulatory mandate requires tech companies to label content that has been generated or edited by artificial intelligence in a manner identifiable to computer systems. The implementation of these digital markers has triggered significant discontent on social media platforms, particularly on Reddit, where users argue the system disproportionately affects individuals using the tool for work or academic purposes.
Critics have characterised the policy as draconian and unethical, with some describing the watermarks as a "digital tattoo" that will catch average users while savvy individuals can bypass detection. One Reddit user, identified as visionode, argued that the system targets students reorganising paragraphs, journalists summarising transcripts, and writers seeking synonyms. This user claimed that while advanced users could hide evidence of AI usage by paraphrasing outputs through other services, ordinary users would be unfairly flagged.
Other users defended the watermarks as a necessary measure to detect AI-generated material and mitigate associated risks. One poster stated there was "literally no good argument" against the policy, suggesting that opposition stems from a desire to deceive others. Another user noted that the watermarks allow for the identification of algorithm-generated content, which is essential for managing the risks posed by AI outputs in various situations.
The debate has also highlighted concerns regarding hypocrisy, with some critics pointing out that frontier models are trained on other people’s work. One user described the direction as "sinister," noting the irony of an AI that watermarks user work given how its own training data was acquired. Another user questioned what the system was claiming credit for, arguing that they had done the lion’s share of the work by providing instructions, context, and refinements.
Amidst the backlash, some users pointed out inconsistencies in the complaints, with one quipping that a critic could not even write their complaint about the watermarks without using Claude. The incident occurs against a backdrop of broader industry shifts, including Twitch allowing streamers to opt out of Amazon’s AI training data and Google’s Gemini app reaching one billion monthly active users in August 2026.
