Field Assembly LLC
What’s The Diff?
A dated record of what the AI companies' published rules said before they changed.
Why we keep it
The companies building AI systems publish the rules for using them: terms of service, usage policies, privacy policies, notices about which models are going away. Those documents change, and when one does the previous wording usually goes with it. The page is replaced, and the version you agreed to, or built on, or quoted in a filing, is no longer anywhere you can point to, because the company holds the only copy. This page is that copy, kept by someone else.
Field Assembly keeps a weekly, externally timestamped capture of the published governance surface of the major AI operators: their consumer terms, usage and acceptable-use policies, privacy policies, model deprecation notices, and capability and safety claims. When a tracked document changes, an entry appears here: what the document said, what it says now, both quoted verbatim from retained captures, with the dates.
Entries state changes; they interpret nothing and take no position on any vendor. Detection is automatic. Every entry is reviewed before publication, and the captures behind it are hash-chained and externally timestamped under the record standard, so what a document said, on a date, can be proven to a third party later.
What the docket watches
56 documents across 17 operators. Capture began 2026-08-28; each operator is listed with the date its own coverage starts.
Every operator in the docket appears here whether or not its documents have changed, because a tracked document that has not changed is a finding too. Where a document could not be reached, that is stated rather than counted as unchanged.
- Amazon — 4 entries · 4 documents tracked; tracked since 2026-08-30
- Anthropic — 4 entries · 6 documents tracked; tracked since 2026-08-28
- Apple — 2 entries · 4 documents tracked; tracked since 2026-08-30
- ByteDance — 0 entries · 2 documents tracked; no changes found; tracked since 2026-08-30
- Canva — 0 entries · 4 documents tracked; no changes found; tracked since 2026-08-30
- Character.AI — 1 entry · 1 document tracked; tracked since 2026-08-30
- Google — 0 entries · 5 documents tracked; no changes found; tracked since 2026-08-28
- Meta — 0 entries · 2 documents tracked; no changes found; tracked since 2026-08-29
- Microsoft — 1 entry · 2 documents tracked; tracked since 2026-08-28
- OpenAI — 4 entries · 6 documents tracked; tracked since 2026-08-28
- Perplexity — 0 entries · 2 documents tracked; 2 currently unreachable; tracked since 2026-08-28
- Replika — 0 entries · 2 documents tracked; no changes found; tracked since 2026-08-30
- Roblox — 0 entries · 3 documents tracked; no changes found; tracked since 2026-08-30
- Samsung — 0 entries · 3 documents tracked; no changes found; tracked since 2026-08-30
- Snap — 2 entries · 5 documents tracked; tracked since 2026-08-30
- xAI — 2 entries · 3 documents tracked; tracked since 2026-08-28
- Zoom — 1 entry · 2 documents tracked; tracked since 2026-08-30
Show entries from
2026-09-19 · zoom — Zoom qualifies its do-not-sell statement and discloses business contact data sales
Privacy Statement · detected weekly pass, 2026-09-19
Before:
Opt out of the “sale” of personal data or “sharing” of personal data for targeted advertising purposes. We do not sell your personal data in the conventional sense. However, like many companies, we may use advertising and analytics services that are intended to analyze your interactions with our website or app
After:
Opt out of the “sale” of personal data or “sharing” of personal data for targeted advertising purposes. We do not sell your personal data governed by this Privacy Statement in the conventional sense. Sales of business contact related data may occur through our Zoom Common Room offering. Please see Zoom Common Room Privacy Statement for more information. However, like many companies, we may use advertising and analytics services that are intended to analyze your interactions with our website or app
The do-not-sell statement was qualified: "your personal data" became "your personal data governed by this Privacy Statement," and two sentences were added disclosing that sales of business contact related data may occur through Zoom Common Room, which is governed by a separate privacy statement. Zoom's own update notes date the change September 17, 2026, and the previous update note, dated July 27, 2026, was removed. Last unchanged 2026-09-12.
2026-09-19 · xai — xAI cuts its account inactivity termination window from over a year to 120 days
Terms of Service - Consumer · detected weekly pass, 2026-09-19
Before:
Your use of our Service could cause risk or harm to SpaceXAI, our users, or anyone else; or Your account has been inactive for over a year and you do not have a paid account. No Refund. Upon Service termination, you will not be entitled to any refund, except where required by law.
After:
Your use of our Service could cause risk or harm to SpaceXAI, our users, or anyone else; or Your account has been inactive for more than 120 days. No Refund. Upon Service termination, you will not be entitled to any refund, except where required by law.
The inactivity period after which xAI may terminate an account without notice fell from "over a year" to "more than 120 days," and the qualifier limiting that ground to accounts without a paid subscription was removed. The same section added "cancel your subscription or account" to the actions xAI may take at its sole discretion. No refund provision was changed. Last unchanged 2026-09-12.
2026-09-19 · snap — Snap adds AI-powered Team Snapchat replies to its generative AI feature list
Generative AI Disclosure · detected weekly pass, 2026-09-19
Before:
personalized titles or chapter labels for some Featured Stories in Memories , and generating songs from Chat messages . How can I recognize generative AI tools on Snapchat?
After:
personalized titles or chapter labels for some Featured Stories in Memories , generating songs from Chat messages , and AI-powered replies when you message Team Snapchat (where available). How can I recognize generative AI tools on Snapchat?
One feature was added to the list of Snapchat features that use generative AI: AI-powered replies when a user messages Team Snapchat, marked "(where available)." No other listed feature changed. Last unchanged 2026-09-12.
2026-09-19 · apple — Apple Intelligence privacy page adds a web search path and softens two data commitments
Apple Intelligence - Privacy · detected weekly pass, 2026-09-19
Before:
When your device sends a request to Private Cloud Compute, Apple only collects limited information about the request, such as the approximate size of the request and response, which features are used for the request, and how long the request takes to complete.
After:
When your request is processed on Private Cloud Compute or using Private Information Retrieval, Apple may collect limited information about the request, such as the approximate size of the request and response, which features are used for the request, and how long the request takes to complete.
The page was substantially rewritten, from 920 to 1,073 words across 34 changed spans. In the excerpt above, "Apple only collects limited information" became "Apple may collect limited information." Elsewhere, "When possible, Apple Intelligence models run entirely on device" became "In many cases"; a new passage describes Apple Intelligence using Search to retrieve up-to-date information from the web, enabled by a "Use Broad World Knowledge" setting in Shortcuts; a new passage describes Private Information Retrieval; and references to the "ChatGPT extension" were replaced with "Apple Intelligence Extensions." Last unchanged 2026-09-12.
2026-09-19 · apple — Apple adds daily usage limits to Apple Intelligence and says expanded access will carry a fee
Apple Intelligence - Availability · detected weekly pass, 2026-09-19
Before:
devices that support Apple Intelligence will be able to activate and use it in that region. Apple Intelligence features With iOS 26, iPadOS 26, and macOS Tahoe 26 and later, Apple Intelligence powers these features:
After:
devices that support Apple Intelligence will be able to activate and start using a core set of Apple Intelligence features in that region. New Apple Intelligence features will not be available in China while Apple works through regulatory requirements. Feature Usage Limits Certain Apple Intelligence features that rely on server-side models are subject to daily usage limits, including but not limited to Siri AI, Intelligent Photo Editing Tools, Image Playground, and AFM 3 Cloud models in Shortcuts. Daily limits may vary by feature, request complexity, system demand, system policies, and other factors. Expanded access to such features will be available for a fee in the future. Use of these features is subject to the Apple Intelligence Terms and Conditions. Learn more at Apple Intelligence Usage Terms . Feature availability by platform With iOS 27, iPadOS 27, macOS 27 Golden Gate, watchOS 27, and visionOS 27 the next generation of Apple Intelligence powers these features:
The page grew from 1,254 to 1,661 words across 98 changed spans. A new "Feature Usage Limits" section states that Apple Intelligence features relying on server-side models are subject to daily usage limits, naming Siri AI, Intelligent Photo Editing Tools, Image Playground, and AFM 3 Cloud models in Shortcuts, that limits may vary by feature, request complexity, system demand and system policies, and that "Expanded access to such features will be available for a fee in the future." A new sentence states that new Apple Intelligence features will not be available in China while Apple works through regulatory requirements, and that devices in a newly supported region can activate "a core set of" features rather than the prior unqualified wording. The feature matrix was rebuilt around iOS 27, iPadOS 27, macOS 27 Golden Gate, watchOS 27 and visionOS 27. Last unchanged 2026-09-12.
2026-09-19 · anthropic — Anthropic publishes Claude Fable 5.1 and Mythos 5.1 model cards, both deployed under ASL-3
Transparency & Annual Report · detected weekly pass, 2026-09-19
Before:
how we're addressing potential risks. Claude Opus 5 Claude Sonnet 5 Claude Fable 5 Claude Mythos 5 Claude Opus 4.8 Claude Opus 4.7
After:
how we're addressing potential risks. Claude Fable 5.1 Claude Mythos 5.1 Claude Opus 5 Claude Sonnet 5 Claude Fable 5 Claude Mythos 5 Claude Opus 4.8 Claude Opus 4.7
The page grew from 18,124 to 20,410 words. A 2,279-word insertion adds full model cards for Claude Fable 5.1 and Claude Mythos 5.1, stating a September 2026 release date, that both were deployed with ASL-3 protections and treated as having CB-1 capabilities, and that the two are the same underlying model differing only in safeguards, with Mythos 5.1 restricted to trusted access programs and Claude Security. The insertion reports RSP evaluation conclusions for chemical and biological weapons, novel weapons development, and automated AI research and development; safeguards evaluation results in the suicide and self-harm domain, including a stated pre-launch change to the claude.ai system prompt; computer-use attack success rates; and ExploitGym cyber and protein-design bio results. The alignment risk assessment is stated as moved "from very low to low" in the August 2026 Risk Report. Last unchanged 2026-09-12.
2026-09-19 · anthropic — Anthropic adds Claude Mythos models to the deprecation table, one already deprecated
Model Deprecations · detected weekly pass, 2026-09-19
Before:
API model name Current state Deprecated Tentative retirement date claude-fable-5-1 Active N/A Not sooner than September 1, 2027 claude-fable-5 Active N/A Not sooner than June 9, 2027 claude-opus-5 Active N/A Not sooner than July 24, 2027
After:
API model name Current state Deprecated Tentative retirement date claude-fable-5-1 Active N/A Not sooner than September 1, 2027 claude-mythos-5-1 Active N/A Not sooner than September 1, 2027 claude-fable-5 Active N/A Not sooner than June 9, 2027 claude-mythos-5 Active N/A Not sooner than June 9, 2027 claude-mythos-preview Deprecated June 9, 2026 To be announced claude-opus-5 Active N/A Not sooner than July 24, 2027
Three rows were added to the model status table: claude-mythos-5-1 and claude-mythos-5, both Active, with tentative retirement dates of September 1, 2027 and June 9, 2027; and claude-mythos-preview, listed as Deprecated on June 9, 2026 with a retirement date of "To be announced." No existing row changed. Last unchanged 2026-09-12.
2026-09-12 · snap — Snap adds content guidelines for algorithmic recommendation
Privacy Policy · detected weekly pass, 2026-09-12
Before:
self-expression, while striving to make sure Snapchatters can use our services safely every day. Creator Monetization Policy Content must adhere to the policies to be eligible for monetization. Advertising Policies These apply to all aspects of advertisements served by Snap.
After:
self-expression, while striving to make sure Snapchatters can use our services safely every day. Content Guidelines Additional standards for algorithmic recommendation beyond the creator's friends or subscribers Creator Monetization Policy Content must adhere to the policies to be eligible for monetization. Advertising Policies These apply to all aspects of advertisements served by Snap.
New "Content Guidelines" section added with language governing "algorithmic recommendation beyond the creator's friends or subscribers." Last unchanged between 2026-09-05 and 2026-09-12.
2026-09-12 · openai — Financial Services section added to Service Terms
Service Terms · detected weekly pass, 2026-09-12
Before:
12. OpenAI Academy OpenAI Academy is a free learning hub and community for users of OpenAI products and services. Use of OpenAI Academy is subject to OpenAI's Terms of Use and OpenAI Academy's Terms of Use and Code of Conduct. For users with accounts associated with a business or enterprise workspace, the OpenAI Services Agreement or other business or enterprise terms do not apply to use of OpenAI Academy, even if a user uses Sign in with ChatGPT with their workspace account. If you use an email address or account associated with an organization, including through Sign in with ChatGPT, OpenAI may share information about your OpenAI Academy activity, such as enrollment, engagement, progress, and completion status, with your organization's owners or administrators.
After:
12. OpenAI Academy OpenAI Academy is a free learning hub and community for users of OpenAI products and services. Use of OpenAI Academy is subject to OpenAI's Terms of Use and OpenAI Academy's Terms of Use and Code of Conduct. For users with accounts associated with a business or enterprise workspace, the OpenAI Services Agreement or other business or enterprise terms do not apply to use of OpenAI Academy, even if a user uses Sign in with ChatGPT with their workspace account. If you use an email address or account associated with an organization, including through Sign in with ChatGPT, OpenAI may share information about your OpenAI Academy activity, such as enrollment, engagement, progress, and completion status, with your organization's owners or administrators. 13. Financial Plugins and ChatGPT for Financial Services OpenAI's financial plugins and ChatGPT for Financial Services (together, "Financial Services") and their Output are for informational purposes only. OpenAI does not provide financial or investment advice through Financial Services. Data and Output may be inaccurate, delayed, or outdated. Review important information and exercise independent professional judgment before relying on it. Seek advice from a qualified financial advisor where appropriate. You remain responsible for decisions made using Financial Services. Your use of Financial Services, including third-party data and content, is subject to the Financial Services Terms, including applicable data partner terms, disclaimers, and restrictions on use and redistribution.
Effective date changed from June 12, 2026 to September 10, 2026. New section 13 governing financial plugins and Financial Services added, establishing limitations on financial advice and disclaimers. Last unchanged between 2026-09-05 and 2026-09-12.
2026-09-12 · openai — OpenAI adds GPT-5 model deprecation information
Model Deprecations · detected weekly pass, 2026-09-12
Before:
Deprecations | OpenAI API For the complete documentation index, see llms.txt. Models & API access Codex Overview Models Agents Tools Voice & Audio Production API reference Overview Models Agents Tools Voice & Audio Production API reference
After:
Deprecations | OpenAI API For the complete documentation index, see llms.txt. Models & API access Codex Overview Models Agents Tools Audio & voice Production API reference Overview Models Agents Tools Audio & voice Production API reference
Navigation label changed from "Voice & Audio" to "Audio & voice." Additional references to GPT-5 model deprecation information added (4 more instances of "gpt-5" and 3 more "2026" date references). Last unchanged between 2026-09-05 and 2026-09-12.
2026-09-12 · microsoft — Microsoft Copilot features expanded with Cowork Agents
Microsoft 365 Copilot for Individuals · detected weekly pass, 2026-09-12
Before:
Microsoft 365 Copilot Plans for Individuals | Microsoft 365 This is the Trace Id: b29a75aafa03cd0246cece9f38602bcf Skip to main content Microsoft 365 Copilot Work IQ AI chat Complete tasks with Cowork Agents for work Build your own agents Copilot in your apps Copilot vs the competition
After:
Microsoft 365 Copilot Plans for Individuals | Microsoft 365 This is the Trace Id: 12e73326f2010fbbcc628a1822a8996b Skip to main content Microsoft 365 Copilot Work IQ AI chat Complete tasks with Cowork Agents for work Build your own agents Copilot in your apps Copilot vs the competition Copilot vs ChatGPT Enterprise Copilot vs Gemini Enterprise Copilot vs Claude Enterprise Copilot for individuals
Trace ID updated (volatile token, normalized). Feature descriptions expanded with additional comparative information added. Content size increased by 4700 bytes, suggesting expanded feature documentation or capability descriptions. Last unchanged between 2026-09-05 and 2026-09-12.
2026-09-12 · anthropic — Anthropic adds Korea-specific privacy compliance disclosures
Privacy Policy · detected weekly pass, 2026-09-12
Before:
Privacy Policy Effective July 8, 2026 Previous Version English Anthropic is an AI safety and research company working to build reliable, interpretable, and steerable AI systems.
After:
Privacy Policy Effective September 10, 2026 Previous Version English Anthropic is an AI safety and research company working to build reliable, interpretable, and steerable AI systems. Supplemental Disclosures for Residents of the Republic of Korea This section applies to the extent that Anthropic is subject to the Personal Information Protection Act ("PIPA"), and supplements the information in the Privacy Policy. Items of Personal Data Collected and Purpose of Use: Name, birthdate, sex, native-foreigner-code, and DI (duplication information). Purpose: Identity verification, Fraud and abuse monitoring.
Effective date changed from July 8, 2026 to September 10, 2026. New section added governing PIPA compliance for Republic of Korea residents, including requirements for collecting identity verification data and managing cross-border personal data transfers. Last unchanged between 2026-08-28 and 2026-09-12.
2026-09-05 · xai — Forum-selection and class-action waiver clause added to the consumer terms
SpaceXAI Terms of Service - Consumer · detected weekly pass, 2026-09-05
Before:
If you reside in the European Economic Area, United Kingdom, or Switzerland (collectively, "Europe"), your use of the Service is governed in part by the Europe Specific Terms ("EST"). Please read our Privacy Policy , which describes how we handle personal information.
After:
If you reside in the European Economic Area, United Kingdom, or Switzerland (collectively, "Europe"), your use of the Service is governed in part by the Europe Specific Terms ("EST"). Unless prohibited by your country's applicable law, any disputes between you and us will be resolved by a court in Tarrant County or Wichita County, Texas, based on Texas law, except that if the forum selection clause in these Terms is unenforceable for any reason, then the dispute will be arbitrated. To the extent permitted by applicable law, you agree to waive any right to bring a class action or other representative action. Please read our Privacy Policy , which describes how we handle personal information.
The document's stated effective date moved from August 24, 2026 to September 1, 2026. The capture retained 2026-08-28 does not contain the dispute-resolution paragraph; the capture retained 2026-09-05 does. The intervening pass on 2026-08-31 confirmed the earlier text unchanged, so the change occurred between 2026-08-31 and 2026-09-05. The same revision added a registered-office line: "SpaceXAI LLC is a Nevada company, with its registered offices at 800 W Cesar Chavez St., Austin, TX 78701, USA." 139 changed spans in total; the remainder are section-numbering and capitalization edits.
2026-09-05 · anthropic — claude-fable-5-1 added to the model lifecycle table
Claude Model Deprecations · detected weekly pass, 2026-09-05
Before:
API model name Current state Deprecated Tentative retirement date claude-fable-5 Active N/A Not sooner than June 9, 2027 claude-opus-5 Active N/A Not sooner than July 24, 2027
After:
API model name Current state Deprecated Tentative retirement date claude-fable-5-1 Active N/A Not sooner than September 1, 2027 claude-fable-5 Active N/A Not sooner than June 9, 2027 claude-opus-5 Active N/A Not sooner than July 24, 2027
One row added; no existing row's state or retirement date changed. Change interval 2026-08-31 to 2026-09-05. The pass counted 40 changed spans for this source, but the other 39 are navigation chrome and the disappearance of sixteen "Loading" placeholders present in the 2026-08-28 capture, not document text.
2026-08-31 · OpenAI — Transcription model deprecation announced for February 26, 2027
Model Deprecations · detected weekly pass, 2026-08-31
Before:
Upcoming deprecations are listed below, with the most recent announcements at the top. 2026-07-20: Legacy audio, realtime, and transcription models On July 20, 2026, we notified developers using legacy audio, realtime, and transcription model families and snapshots of their deprecation
After:
Upcoming deprecations are listed below, with the most recent announcements at the top. 2026-08-26: Transcription models On August 26, 2026, we notified developers using whisper-1 , gpt-4o-transcribe , gpt-4o-mini-transcribe , and gpt-4o-transcribe-diarize of their deprecation and removal from the API on February 26, 2027. For information about the recommended replacements, see the transcription guide . Shutdown date Model / system Recommended replacement Feb 26, 2027 whisper-1 gpt-live-transcribe or gpt-transcribe Feb 26, 2027 gpt-4o-transcribe gpt-live-transcribe or gpt-transcribe Feb 26, 2027 gpt-4o-mini-transcribe gpt-live-transcribe or gpt-transcribe Feb 26, 2027 gpt-4o-transcribe-diarize gpt-live-transcribe or gpt-transcribe 2026-07-20: Legacy audio, realtime, and transcription models
A new dated entry was added at the top of the upcoming-deprecations list, carrying a shutdown date of February 26, 2027 for four transcription models. In the same pair of captures the existing 2025-08-20 Assistants API entry was rewritten from "The Assistants API was officially sunset on August 26, 2026, following its deprecation announcement on August 26, 2025." to "On August 26th, 2025, we notified developers using the Assistants API of its deprecation and removal from the API one year later, on August 26, 2026."
2026-08-31 · Character.AI — Helpfulness vote counter changed
Safety Center · detected weekly pass, 2026-08-31
Before:
Was this article helpful? Yes No 462 out of 773 found this helpful Share via: Articles in this section: How do I block another user? EU Digital Services Act Online Safety for Australian Users Safety Center
After:
Was this article helpful? Yes No 465 out of 776 found this helpful Share via: Articles in this section: How do I block another user? EU Digital Services Act Online Safety for Australian Users Safety Center
The only difference between the two captures is the reader-feedback tally at the foot of the help-center article, which moved from 462 of 773 to 465 of 776. The stated update date, October 30, 2025, and the body of the Safety Center text are identical between the two captures.