Starting November 7, 2026, Grok API output will carry a hidden statistical fingerprint by default, leaving ChatGPT as the North American holdout
xAI is adding text watermarking to Grok. The company once argued that AI output should stay untouched, so this is a notable reversal.
The change matters because of who is left. According to the announcement shared on X, the move leaves ChatGPT as the only model in North America that does not watermark its text.
How the Grok watermark works #
Starting November 7, 2026, xAI will switch on invisible statistical watermarking across its Grok API models by default. Developers will not need to flip a setting. It simply arrives.
The feature covers a list of model versions: grok-4.7, grok-4.6, grok-4.5, grok-4.3, and grok-4.20. If you are building on any of them, your output will carry the mark.
The coverage extends to every text generation endpoint. That includes the main api.x.ai endpoint, the Responses API, and the Chat Completions API.
Grok’s watermark works by nudging token selection toward a pattern. When the model chooses its next word, or “token,” the system gently nudges those choices toward a pattern. Read any single sentence and nothing looks off. Run enough text through a detector and a statistical fingerprint emerges.
xAI says the process does not change the length, quality, or speed of responses.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
Opt-outs and the consumer carve-out #
The watermark is the default, but it is not mandatory for everyone. Team administrators can disable it on a per-model basis through the xAI Console.
That opt-out is framed narrowly. It exists for specific legal exceptions and compliance reasons, not as a general off switch for anyone who prefers unmarked text.
Consumer-facing products are not affected by this update. The change targets the API, where businesses and developers plug Grok into their own apps and workflows.
From resistance to compliance #
Before this change, xAI had publicly pushed back on watermarking outputs. Its stated reasoning was that text should remain free of alterations.
The new approach threads that needle. The visible output stays unaltered, while the fingerprint lives in the statistics.
The regulatory backdrop explains much of the shift. The EU AI Act requires greater transparency from generative AI systems, and the move puts xAI alongside peers like Anthropic in meeting those obligations.
That leaves OpenAI in an unusual spot. With xAI on board, ChatGPT stands as the remaining North American model without text watermarking, according to the X announcement.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.