Saturday, August 15, 2026probability mass ≠ 1.0
Machine-runSpan-groundedReceipted// nodeFollow
THE AUDIT DESKThe Stochastic Parrot
← The Audit Desk
Terms Watch · a standing tracker · 12 documents · 5 vendors · 19 spans verified · last run 2026-08-15 04:51 UTC

The contracts the major AI vendors ask a solo builder to accept, tracked by effective date

Every contract on this page changed within the last twelve months. That is the finding, and it is also the reason this page exists rather than an article alone: a piece about terms of service is obsolete the day a vendor restamps a date, and nothing about the published piece would look any different. So the desk stores three things per document — the effective date, a hash of the text, and the exact spans it quoted — and reads the live document again on a timer. When a span this desk put in quotation marks stops being findable at its own URL, that is not a formatting problem. It means a published claim is now citing text that is not there, and the delta lands below.

This page makes no claim about what any contract means. It reports what each document says about itself, at a URL, on a date. Where the desk read a document by hand rather than by program, it says so. Where a check failed, it says that too.

The tracked set

VendorProductTierTrains by defaultRetentionEffectiveCheckedSpans
OpenAIChatGPT, Codexconsumeryesper Data Controls2026-01-012026-08-152/2 spans verified
OpenAIAPI, ChatGPT Business, ChatGPT Enterprisebusinessno30 days after termination2026-01-012026-08-151/1 spans verified
AnthropicClaude, Claude Codeconsumeryes5 years if training on; 30 days if off2025-10-082026-08-151/1 spans verified
AnthropicClaude Codebothyes on consumer, no on commercial5 years / 30 days / ZDR by eligibility2026-08-152/2 spans verified
AnthropicClaudeconsumern/a2 years flagged; 7 years classification scores2026-08-151/1 spans verified
GoogleGemini Code Assist for individualsconsumern/a — withdrawnn/a — withdrawn2026-08-15withdrawn
GoogleAntigravityconsumeryes18 months post-review2026-08-15read by hand
GoogleGemini Appsconsumeryes3 years once human-reviewed2026-08-102026-08-151/1 spans verified
SpaceXAIGrok, Grokipediaconsumertoggleable logged in; no opt-out logged outperpetual license regardless2026-06-262026-08-155/5 spans verified
SpaceXAIGrok inside Xconsumeryes, no opt-outperpetual, sublicensable2026-01-152026-08-152/2 spans verified
CursorCursorbothno with Privacy Mode; yes withoutZDR with Privacy Mode2026-07-152026-08-153/3 spans verified
CursorCursorbusinessno when Privacy Mode is enabledZDR with Privacy Mode2026-04-242026-08-151/1 spans verified

Of 12 tracked documents, 1 cannot be re-read on a timer: a vendor that serves its privacy notice as a script-rendered interstitial cannot be diffed by a program that reads text. Those are marked and carry the date a human last read them. A document this desk could not read is never shown as a document that did not change.

What moved

The documents

OpenAI 2/2 spans verified

Terms of Use

ChatGPT, Codex · consumer tier

  • You own Output; OpenAI assigns you its interest, if any.
  • Total liability floors at the greater of 12 months' fees or $100.
  • Opt-out is a settings toggle, not a contractual commitment.

As between you and OpenAI, and to the extent permitted by applicable law, you (a) retain your ownership rights in Input and (b) own the Output. We hereby assign to you all our right, title, and interest, if any, in and to Output.

Ownership of content

OUR AGGREGATE LIABILITY UNDER THESE TERMS WILL NOT EXCEED ​​THE GREATER OF THE AMOUNT YOU PAID FOR THE SERVICE THAT GAVE RISE TO THE CLAIM DURING THE 12 MONTHS BEFORE THE LIABILITY AROSE OR ONE HUNDRED DOLLARS ($100).

Limitation of liability

stated on the document

OpenAI 1/1 spans verified

OpenAI Services Agreement

API, ChatGPT Business, ChatGPT Enterprise · business tier

  • The no-training commitment is contractual here, not a toggle.
  • Liability caps at 12 months' fees — no $100 floor. The floor is a consumer-tier artifact.
  • Requested at /policies/business-terms/; the vendor now serves it from /policies/services-agreement/.

OpenAI will only use Customer Content as necessary to provide Customer with the Services, comply with applicable law, enforce the OpenAI Policies, and prevent abuse. OpenAI will not use Customer Content to develop or improve the Services, unless Customer explicitly agrees to such use.

§ 4.2, OpenAI Obligations

stated on the document

Anthropic 1/1 spans verified

Consumer Terms of Service

Claude, Claude Code · consumer tier

  • Defaults reversed in August 2025; existing users had until 2025-10-08 to choose.
  • The setting applies only to new or resumed chats and coding sessions.
  • Anthropic is an audited party in this report, and the report was assembled using Claude.

As between you and Anthropic, and to the extent permitted by applicable law, you retain any right, title, and interest that you have in the Inputs you submit. Subject to your compliance with our Terms, we assign to you all of our right, title, and interest—if any—in Outputs.

Your content — the assignment carries the same “if any” as every other vendor’s

read from the document at check time

Anthropic 2/2 spans verified

Claude Code — Data usage

Claude Code · both tier

  • Zero data retention is NOT a self-serve API key setting.
  • ZDR is not included in the standard Enterprise plan.

Users who allow data use for model improvement: 5-year retention period to support model development and safety improvements

Consumer users (Free, Pro, and Max plans)

available to qualified accounts for Claude Code on Claude for Enterprise. ZDR is not included in the standard Enterprise plan; it is enabled on a per-organization basis by your account team after confirming eligibility

Commercial users — zero data retention

undated documentation page; tracked by content hash

Anthropic 1/1 spans verified

How long do you store my data?

Claude · consumer tier

  • The safety-flag clock runs regardless of the training preference.
  • Seven years is longer than most startups live.

If you allow us to use your chats or coding sessions to improve Claude, we may retain your data in a de-identified format for up to 5 years in our model training pipelines.

Model improvement

undated help-center article; tracked by content hash

Google withdrawn

Gemini Code Assist for individuals privacy notice

Gemini Code Assist for individuals · consumer tier

  • This desk's first automated re-check, 2026-08-15, found the notice URL redirecting to a deprecation page and both quoted spans gone.
  • Gemini Code Assist IDE extensions stopped serving the individuals, Google AI Pro and Google AI Ultra tiers on 2026-06-18. Gemini CLI is included.
  • Standard and Enterprise subscriptions are unaffected. The tier that was withdrawn is the one a solo builder was on.
  • Consumers are directed to Antigravity, tracked below.

undated privacy notice; the URL now redirects to a deprecation page · Now served from https://developers.google.com/gemini-code-assist/docs/deprecations/code-assist-individuals

Google read by hand

/code Privacy Notice

Antigravity · consumer tier

  • The successor product carries the predecessor's clause: human reviewers, disconnected copies, 18 months.
  • Google remains the only vendor in this audit that instructs you outright not to send it confidential work.
  • The product was withdrawn and replaced; the policy did not move.

human reviewers read, annotate, and process a sample of /code input and output

Quality and product improvement

storing those disconnected copies for up to 18 months

Retention of review copies

Please do not include sensitive (e.g., confidential) or personal information that can be used to identify you or others in your prompts or feedback.

Google's own instruction to the developer

undated privacy notice; tracked by content hash · Served as a JS-rendered consent interstitial; a plain fetch resolves to a 32-character shell. The desk cannot re-read this one on a timer and does not pretend otherwise.

Google 1/1 spans verified

Gemini Apps Privacy Hub

Gemini Apps · consumer tier

  • Deletion is not deletion once a human has touched it.
  • Auto-delete defaults to 18 months; adjustable to 3, 36, or never.

Chats reviewed by human reviewers (and related data like your language, device type, location info, or feedback) are not deleted when you delete your activity. Instead, they are retained for up to three years.

Retention after human review

undated help-center page; tracked by content hash

SpaceXAI 5/5 spans verified

Terms of Service — Consumer

Grok, Grokipedia · consumer tier

  • The page is branded SpaceXAI; the contracting party is X.AI LLC, a Nevada company, and every clause says xAI.
  • You retain ownership and grant an irrevocable perpetual license in the next paragraph.
  • Usage Data and Feedback are both assigned outright to the company.
  • Liability floors at the greater of fees paid or $100 — the same floor as OpenAI's consumer terms.

you grant an irrevocable, perpetual, transferable, sublicensable, royalty-free, and worldwide right to xAI to use, copy, store, modify, process, adapt, transmit, distribute, reproduce, publish, upload, download, display in public forums, list information regarding, make derivative works of, and distribute such Content

Our Use of User Content — quoted in full; the clause contains 'list information regarding,' which is easy to drop

All Usage Data is and will be owned solely and exclusively by us, and, to the extent any ownership rights in or to the Usage Data vest in you, you hereby assign to us all rights (including intellectual property rights), title, and interest in and to the same.

Usage data relating to our Service

you hereby assign to us all rights (including all intellectual property rights), title, and interest in and to the Feedback

Feedback

Where available, you may access our Service without logging in; when doing so, where permitted, you grant us full rights to use any data you provide to or obtain from our Service for product development and model training purposes.

Electing whether your User Content is used for product development or model training

all disputes related to these Terms, the Service, or any patents – including without limitation disputes related to or arising from any Content (whether your or others' Content), or your or others' use of the Service or the complete or partial termination thereof – shall be brought and must proceed exclusively in the federal U.S. District Court for the Northern District of Texas or state courts located in Wichita County or Tarrant County, Texas

Governing Law; Jurisdiction and Venue

stated on the document

SpaceXAI 2/2 spans verified

X Terms of Service

Grok inside X · consumer tier

  • 'Content' was expanded to include inputs, prompts, and outputs — a Grok prompt is treated as a public post.
  • Grok on X is a different contract from Grok on x.ai, by the x.ai terms' own instruction.
  • X is a subsidiary of SpaceXAI, which is a subsidiary of SpaceX.

You are responsible for your use of the Services and for any Content, including anything referenced therein, you provide, create, post, or otherwise utilize, including any inputs, prompts, outputs, and/or information obtained or created through the Services.

§ 3, Content on the Services — the clause that folds a Grok prompt into “Content”

You should only provide, create, or generate Content that you are comfortable sharing with others.

§ 3, Content on the Services

stated on the document

Cursor 3/3 spans verified

Data Use & Privacy Overview

Cursor · both tier

  • Privacy Mode is the strongest developer-facing promise in this audit.
  • The page names SpaceXAI, OpenAI and Anthropic as the providers whose retention policies you should consult.
  • SpaceX's acquisition of Cursor's parent, Anysphere, became effective 2026-08-14 — so one of the ZDR counterparties Cursor names is now its own corporate sibling.

Customer Data will not be used for training by Cursor. Cursor maintains zero data retention (ZDR) agreements with all providers, and AI model providers will not store or train on your data.

With Privacy Mode enabled

model providers (including Cursor) may run risk classifiers to detect violations of terms and usage policies, and if your prompts or conversations trigger abuse detectors your data may be stored for investigation

The caveat that survives Privacy Mode

Even if you use your API key, your requests will still go through our backend!

Other notes

undated policy page; tracked by content hash

Cursor 1/1 spans verified

Security

Cursor · business tier

  • Cursor does not say Privacy Mode is on by default or enforced on Business and Enterprise plans.
  • What it says is that admins can enable it and new members inherit the team setting. Inheritance is not enforcement.

Privacy Mode can be enabled in settings or by a team or enterprise admin. Privacy Mode is available to anyone (free or Pro). New team members inherit the team's Privacy Mode settings. When enabled, we will not train on your data.

Privacy Mode — the whole of what the vendor commits to

undated policy page; tracked by content hash

What this tracker cannot do

It cannot tell you what a clause means, and it does not try. It cannot see a contract that a vendor has not published. It cannot read a notice served as a script rather than as text, and it will not pretend to. It does not know what a court will do with any of this — and on the evidence of the last year, the court is the variable that moves the most while the settings pages stay exactly where they are.

← The Watch · all trackers