# Anthropic expands Claude watermarking to older models globally > Anthropic is expanding its text watermarking across older Claude models, including Opus 4.6 and Sonnet 4.6, with no opt-out for users. Only EU regulators, law enforcement, journalists, and researchers can access the detection tools. - Type: AI news - Page: https://completeaitraining.com/news/anthropic-expands-claude-watermarking-to-older-models/ - Published: 2026-11-04 - Updated: 2026-11-04 - Author: Jeroen Erne - Jobs: Creatives, Customer support, Executives and strategy, Government, Insurance, Legal, Marketing, PR and communications, Sales, Writers - Reading time: 2 min Anthropic is expanding its text watermarking system across older Claude models, including Mythos 5, Opus 4.7, Opus 4.6, and Sonnet 4.6. The move means more text generated by the AI assistant will carry an invisible pattern that automated systems can detect - and users cannot opt out. The watermarking was first added to Anthropic's newest models in August, following EU regulations that require such technology in generative AI tools. Despite being triggered by European law, the company is applying the change globally for all users. #### Who can actually detect the watermark Normal users have no way to check whether their text would be flagged. Detection tools are restricted to eligible organizations under EU law, including regulators, law enforcement, journalists, and researchers. Companies required to verify watermarking under the EU's AI Act also have access. Anthropic has acknowledged weaknesses in the system. Shorter texts may not contain enough of the pattern to be reliably identified. The company has pushed back, however, against claims that watermarking degrades output quality. #### Criticism and workarounds The August announcement drew backlash from users who argued the system was unfair. Critics pointed out that even using Claude for proofreading or fact-checking could trigger the watermark. Others noted that bad actors could simply rephrase watermarked text with another AI tool to strip the markings. Some escape routes remain. On Opus 4.6 and Sonnet 4.6, adjusting the settings that control how the system selects words - accessible through the API - removes the watermark. Users can also switch to even older models, dating back to Opus 3 from 2023, which have not been updated with the watermarking changes. #### How other companies are responding Anthropic stands apart in applying the watermark globally rather than limiting it to EU jurisdictions. OpenAI has introduced a similar feature called textGrain for ChatGPT, but it will only be added automatically for users in Europe. The EU's AI Act continues to shape how generative AI tools handle transparency, and professionals who rely on these systems for drafting, editing, or customer-facing content will need to understand where the invisible markers apply - and where they don't. #### Why this matters for writers, marketers, and support teams If you use Claude to draft client emails, marketing copy, or support responses, that text now carries a detectable watermark - even if you only used the tool for light editing. There is no user-facing checker to verify this yourself. For teams operating under regulatory scrutiny or client confidentiality requirements, switching to older unmarked models or adjusting API settings may be the only way to avoid leaving a trace. Understanding which models are affected and what detection means under EU law is now part of the workflow for anyone producing text with AI. --- This is the Markdown form of https://completeaitraining.com/news/anthropic-expands-claude-watermarking-to-older-models/. The whole site is mapped at https://completeaitraining.com/llms.txt.