Anthropic will begin applying an invisible text watermark to outputs from four more of its Claude models from Tuesday, extending a system that its newer models already carry.
The company says the mark cannot identify the person who used the tool.
Which Models Are Covered
According to an email sent to customers of the Claude Platform, the watermark will apply to Claude Mythos 5, Claude Opus 4.7, Claude Opus 4.6 and Claude Sonnet 4.6.
-
Indian Markets Turning Corner After Two Tough Years, Says Sandeep Tandon -
NDTV Profit Ignite 2026: Experts See Opportunity In Equities, Stay Bullish For Long-Term -
Buy Right, Sit Tight Does Not Work for Individual Stocks, Says Dinshaw Irani -
'Worst Is Behind': Quant MF Sees Market Bottom, Flags 'Extreme Euphoria' In Midcaps -
India's Mutual Fund Industry Could Become A $1 Trillion Opportunity, Say Fund Managers -
'Foreign Investors Are Baraatis': Raamdeo Agrawal's Take On LTCG Tax -
Stay Invested Through Market Cycles: Raamdeo Agrawal at NDTV Profit Ignite -
Ramesh Damani Remains Bullish On India For Next 20-30 Years, Sees Bargains Amid Market Weakness -
CAS Came After A Long Consultation Process: SEBI Chairman Tuhin Kanta Pandey AT NDTV PROFIT IGNITE -
Aim To Cut FPI Onboarding Time From Mth To Five Days: SEBI Chairman Tuhin Kanta Pandey AT NDTV PROFI
The rollout on Amazon Bedrock may take a few additional days. The email says newer models, including Claude Opus 5.5, already carry the same watermark.
How The Watermark Works
The email describes it as an imperceptible pattern in word choice. Anthropic has said its approach is based on SynthID-Text, a method published by Google DeepMind. Over a longer piece of text, the word choices create a statistical pattern that a detector holding the right key can test for. The company says it adds no characters or tokens and does not alter meaning, quality or readability.
It also has no effect on pricing or request formats and only a negligible effect on speed. Code generally carries less watermarking than ordinary prose, because there is less flexibility in word choice.
ALSO READ | AI Race Heats Up: DeepSeek Drops An Autonomous 'Super-Agent' To Take On Claude
What It Does Not Reveal
A viral X post by Pete Sena described the move as Claude "snitching". Anthropic's notice says otherwise: the watermark contains no information about the user, their organisation or their conversations with Claude.
Because the mark sits in the text itself, it can travel with copied and pasted text and may survive some editing. But it is not proof of authorship. Anthropic has said a watermark can show only that Claude was likely involved at some point, and cannot separate "Claude wrote this" from "Claude heavily edited this".
Text a person wrote and merely had proofread, translated or summarised by Claude could therefore carry the mark.
Why Anthropic Is Doing It
The company has said it is adding watermarking to comply with the EU AI Act, and that other developers who signed the same Code of Practice are also expected to follow.
Anthropic has said it cannot yet restrict the feature to specific regions, so it is being deployed globally. For images and other supported files, Claude will attach content credentials through C2PA metadata instead.
ALSO READ | Anthropic IPO: Alphabet Set To Become Largest Shareholder, Amazon Among Biggest Backers
Who Can Check For It
Sena's post claimed that only regulators, journalists, fact-checkers and researchers can currently test text for the mark. That claim could not be independently verified. Anthropic has said it will offer a watermark detection API, though it has not yet provided implementation details.
What Customers Need To Do
Anthropic's email tells customers they need not change anything unless they rely on the watermark for their own compliance and use non-default sampling settings.
Essential Business Intelligence, Sharp Market Insights, Practical Personal Finance Advice, Daily Fuel, Gold and Silver Prices and Latest Stories — On NDTV Profit.