AI News · 4 min read
Grok API watermarks text by default from Nov 7
xAI SpaceXSI turns invisible text watermarking on by default Nov 7, 2026 for seven Grok models — EU AI Act compliance, x-watermark header, admin opt-out only.
What SpaceXSI turned on today (November 7, 2026)
xAI’s SpaceXSI API docs for Output Watermarking ↗ state that invisible text watermarking is on by default starting November 7, 2026 for EU AI Act compliance. It applies on every SpaceXSI API endpoint, including the global https://api.x.ai endpoint and regional endpoints.
Models covered at launch (seven):
grok-4.7
grok-4.6
grok-4.5
grok-4.3
grok-4.20-reasoning
grok-4.20-non-reasoning
grok-4.20-multi-agent
Watermarking covers text from the Responses API and the Chat Completions API — ordinary text, structured outputs, and tool-call arguments. There is no request parameter. You do not need a code change for generation to keep working. To see whether a given response was watermarked, read the response header: x-watermark: true means it was; any other value or a missing header means it was not.
Per the docs, the mark is statistical: it slightly shapes which tokens the model picks. It adds no characters, markup, or hidden text. SpaceXSI’s FAQ claim (vendor statement — attribute it) is that watermarking does not change response quality or speed.
Detection is also statistical: watermarked / not watermarked / uncertain, with longer text giving a stronger signal. Access to the detection API is by application only through a form linked from the docs.
Honest scope note: the official watermarking page is about API text output. It does not clearly say the consumer Grok app or posts on X are watermarked the same way. Do not assume consumer chat or X posts are covered unless xAI publishes that separately.
Adjacent Grok product news (different surface): our earlier desk note on Grok Bot US outbound phone calls.
Secondary coverage of the same default-on change: The Tech Signal ↗, TeslaNorth ↗, and Real Python’s AI news weekly (Nov 6) ↗.
Opt-out (team admins only)
Team admins can turn watermarking off for some or all models in the SpaceXSI Console. The setting applies to every API key on the team. Changes apply to new requests within about five minutes, and every opt-in / opt-out is recorded in the team audit log.
Opting out requires confirming that a legal exception under applicable law applies to your use, and that you are solely responsible for any marking, disclosure, or other legal requirement. Some models may be locked on for a team and cannot be turned off from the Console.
This is a compliance / provenance control — not a “make Grok better” toggle.
What people are saying on X
Developer checklist framing (today):
“Grok API change today (Nov 7): 7 models watermark text by default.
No request param. Check response header x-watermark. Admins can opt out per model in Console.
Try: log that header on your next call before you assume output is "clean."
#AI #Grok”— @GoChengzi ↗, Nov 7, 2026
Default-on + legal-exception framing:
“#Breaking - Starting tomorrow, @spacexai will watermark all Grok API text by default — invisible marks on 7 models, to comply with the EU AI Act. Admins can only opt out by declaring a legal exception.”
Provenance contrast (statistical watermark vs cryptographic proof — do not conflate):
“Two provenance stories in one week:
Anthropic's FlashProve line: cryptographic proof an output came from a claimed model
Grok API: statistical watermark by default starting Nov 7 (EU AI Act citation)
Useful. Incomplete.
Provenance answers "which system spoke."
Validation”
Our take: who should care
Any team calling Grok via the SpaceXSI / xAI API — especially products that ship into the EU or sell to EU customers — should treat today as a policy go-live, not a model upgrade. Log x-watermark on Responses and Chat Completions now so you know which keys / models are marked before compliance asks.
Compliance / legal / content-provenance teams: review Console opt-out only if counsel confirms a real exception. Opt-out is team-wide per model, audited, and some models may stay locked on. Do not treat “off” as a free switch for marketing copy that still needs disclosure.
Vs Anthropic Claude watermarks / FlashProve: different mechanisms. Grok’s API mark is a statistical token-shaping watermark cited for EU AI Act compliance. FlashProve-style cryptographic proof of origin is a different claim. Useful to track both; do not mash them into one “AI watermark” headline.
Not a quality story. SpaceXSI says quality and speed are unchanged (vendor claim). Your action today is observability (x-watermark) and a counsel-backed opt-out decision — not a prompt rewrite.
For where Grok sits among assistants, see our Grok 4 release review and best ChatGPT alternatives 2026.
AdSense-safe note: we are documenting what xAI’s API docs ship and how teams can verify headers — not selling opt-out advice or any paid xAI plan.