Past sample edition. This is Terms Watch for the week of 12 Oct 2026, as corrected on 15 Oct 2026. It is shown in full so you can see what subscribers get each Monday. It is not this week's edition. AI Reports · Pricing · Corrections

DimDrop AI Reports · Terms Watch

What changed in AI vendor terms this week

Week of 12 Oct 2026: changes first seen from 6 Oct to 12 Oct in the 59 documents we track.

This week in 60 seconds

Google and OpenAI retire 36 models by 23 Oct 2026, starting on 20 Oct 2026 with Google's gemini-2.5-flash, gemini-2.5-flash-lite and gemini-2.5-pro.

36 models retire in the next 14 days

Google and OpenAI retire 36 models by 23 Oct 2026. First, on 20 Oct 2026 (in 8 days): Google's gemini-2.5-flash, gemini-2.5-flash-lite and gemini-2.5-pro.

Who: Engineering and platform teams calling these models.

Do: Move gemini-2.5-flash, gemini-2.5-flash-lite and gemini-2.5-pro to gemini-3.8-flash (the vendor's first named replacement) before 20 Oct 2026; the calendar below lists every model and replacement.

OpenAI: Data controls in the OpenAI platform

The table adds /v1/audio/voices: data is not used for training, abuse monitoring retention is 30 days, application state is kept until deleted, and it is not Zero Data Retention eligible. The page changed 2 more times this week; each change is listed below.

Who: Privacy and compliance teams using OpenAI's API.

Do: Check that your OpenAI retention settings still match your own policy. See “Storage requirements and retention controls per endpoint”.

xAI: Privacy Policy

SpaceXSI no longer asks you to keep personal information out of prompts, extends payment processing to other purchases, and says you are responsible for all actions, including providing payment information.

Who: Privacy and compliance teams using xAI's paid consumer plans.

Do: Compare it with your xAI DPA and your vendor register. See “2. Personal information we collect (Notice at collection)”.

4 matrix cells changed since 5 Oct 2026

For example, Google, Vertex AI (Google Cloud), default retention: was Not stated, now 1 day.

Who: Anyone who keeps a vendor comparison or risk register.

Do: Update your vendor records; each cell below cites its source.

Microsoft: Azure OpenAI model deprecations & retirements

Models can launch directly into general availability, and generally available SpaceXAI Grok models from Grok 4.7 onward get a minimum of six months instead of the standard 18 month lifecycle.

Who: Engineering and platform teams using Microsoft's API.

Do: Note the new notice terms in your Microsoft vendor file. See “Lifecycle stages”.

Who should read the rest: anyone who signs, audits or builds on these vendors' plans. Everything below is grouped and skippable; each item folds its evidence under "How we know".

9material changesact on these

13notable changesworth a look

4matrix cells changedsince 5 Oct

36models retiringin the next 30 days

Contract and privacy check

Topics a Material change touched this week, across the 59 documents we track.

For counsel
IP and indemnity	no material change
Liability	no material change
Governing law and disputes	no material change
Notice of changes	Microsoft
For privacy
Training on your data	no material change
Retention and zero data retention	OpenAI
Human review	no material change
Data residency	OpenAI
Subprocessors and sharing	xAI

Jump to a vendor (this week's changes):

Material changes

Changes a team should act on: what the vendor does with your data, ownership, liability, notice, consent or model dates.

MaterialOpenAIretentionAPI

Data controls in the OpenAI platform

The table adds /v1/audio/voices: data is not used for training, abuse monitoring retention is 30 days, application state is kept until deleted, and it is not Zero Data Retention eligible.

Changed 7 Oct to 8 Oct · confirmed by 4 of 4 models

How we know

The wording under “Storage requirements and retention controls per endpoint” (up to 30 words each):

Before: nothing (new text).

After: No · 30 days · No · Until deleted · /v1/audio/voices

Window: last capture of the old text 7 Oct 2026, first capture of the new text 8 Oct 2026 (1 day).

Sections: Storage requirements and retention controls per endpoint.

Topics: retention.

Graded Material by our published rubric (rules M1) and confirmed by 4 of 4 models reading the changed lines.

Source: Data controls in the OpenAI platform; our capture of 8 Oct 2026, snapshot 3becfa310bfa.

MaterialOpenAIresidencyretentionAPI

Data controls in the OpenAI platform

OpenAI added the /v1/decisions endpoint (gpt-6-luna), available in all API regions with regional processing in the United States and Europe, Zero Data Retention and HIPAA eligibility, and image CSAM scanning.

Changed 6 Oct to 7 Oct · confirmed by 4 of 4 models

How we know

The wording under “/v1/decisions” (up to 30 words each):

Before: nothing (new text).

After: By default, abuse monitoring logs are retained for up to 30 days. Eligible customers can use Zero Data Retention, subject to the limitations below. · Prompt caching may store encrypted ...

Window: last capture of the old text 6 Oct 2026, first capture of the new text 7 Oct 2026 (1 day).

Sections: /v1/decisions; API Endpoint, tool and model support; Support by region; Image and file inputs.

Topics: residency, retention, zero data retention.

Graded Material by our published rubric (rules A1, M1) and confirmed by 4 of 4 models reading the changed lines.

Source: Data controls in the OpenAI platform; our capture of 7 Oct 2026, snapshot f269ae8680ee.

MaterialOpenAIresidencyAPI

Data controls in the OpenAI platform

Fast mode now supports EU data residency for GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna, but not GPT-6 Astra, and GPT-6.1 Sol Ultrafast mode now supports EU data residency.

Changed 8 Oct to 9 Oct · confirmed by 4 of 4 models

How we know

The wording under “API Endpoint, tool and model support” (up to 30 words each):

Before: Fast mode is not available with EU data residency for GPT-6 Astra, GPT-6.1 Sol, GPT-6 Sol, or GPT-6 Luna. GPT-6.1 Sol supports only US and EU data residency. · Fast ...

After: Fast mode supports EU data residency for GPT-6.1 Sol, GPT-6 Sol, and GPT-6 Luna. Fast mode is not available with EU data residency for GPT-6 Astra. GPT-6.1 Sol supports only ...

Window: last capture of the old text 8 Oct 2026, first capture of the new text 9 Oct 2026 (1 day).

Sections: Which models and features are eligible for data residency?; API Endpoint, tool and model support.

Topics: residency.

Graded Material by our published rubric (rules M1) and confirmed by 4 of 4 models reading the changed lines.

Source: Data controls in the OpenAI platform; our capture of 9 Oct 2026, snapshot 2e6b4235f51e.

MaterialxAIdata sharingConsumer paid

Privacy Policy

Changed 11 Oct to 12 Oct · confirmed by 4 of 4 models

How we know

The wording under “2. Personal information we collect (Notice at collection)” (up to 30 words each):

Before: We ask that you do NOT include personal information in your prompts and inputs into our Service; however, we cannot control what you provide to us. · Payment Data: Where ...

After: Payment Data: Where payment is required to access the Service (ex., if you are paying for a subscription), or is otherwise used to make a purchase, we may, or a ...

Window: last capture of the old text 11 Oct 2026, first capture of the new text 12 Oct 2026 (1 day).

Sections: 2. Personal information we collect (Notice at collection).

Topics: data sharing.

Graded Material by our published rubric (rules M1) and confirmed by 4 of 4 models reading the changed lines.

Source: Privacy Policy; our capture of 12 Oct 2026, snapshot 51a60596a215.

MaterialMicrosoftlifecyclenoticeAPI

Azure OpenAI model deprecations & retirements

Models can launch directly into general availability, and generally available SpaceXAI Grok models from Grok 4.7 onward get a minimum of six months instead of the standard 18 month lifecycle.

Changed 6 Oct to 7 Oct

How we know

The wording under “Lifecycle stages” (up to 30 words each):

Before: nothing (new text).

After: Generally available SpaceXAI Grok models starting with Grok 4.7 are available for a minimum of six months. For model-specific retirement dates, see the Model retirement schedule.

Window: last capture of the old text 6 Oct 2026, first capture of the new text 7 Oct 2026 (1 day).

Sections: Microsoft Foundry Models lifecycle and support policy; Lifecycle stages; Generally Available (GA) replacement model overlap commitments; Preview model lifecycle.

Topics: lifecycle, notice.

Source: Azure OpenAI model deprecations & retirements; our capture of 7 Oct 2026, snapshot ef5e9b5a90f7.

MaterialGooglelifecycleAPI

Gemini API deprecations

Google listed gemini-3.8-live with no shutdown date announced and added deep-research-pro-preview-12-2025, shutting down October 23, 2026, replaced by deep-research-preview-04-2026, released April 21, 2026.

Changed 8 Oct to 9 Oct

How we know

The wording under “Managed agents” (up to 30 words each):

Before: Need to tell us more? [[["Easy to understand","easyToUnderstand","thumb-up"],["Solved my problem","solvedMyProblem","thumb-up"],["Other","otherUp","thumb-up"]],[["Missing the information I need","missingTheInformationINeed","thumb-down"],["Too complicated / too many steps","tooComplicatedTooManySteps","thumb-down"],["Out of date","outOfDate","thumb-down"],["Samples / code issue","samplesCodeIssue","thumb-down"],["Other","otherDown","thumb-down"]],["Last updated 2026-10-07 UTC."],[],[]]

After: deep-research-preview-04-2026 · April 21, 2026 · deep-research-max-preview-04-2026 · April 21, 2026 · deep-research-pro-preview-12-2025 · December 11, 2025 · October 23, 2026 · deep-research-preview-04-2026 · Need to tell us more? [[["Easy ...

Window: last capture of the old text 8 Oct 2026, first capture of the new text 9 Oct 2026 (1 day).

Sections: Gemini 3 models; Managed agents.

Topics: lifecycle.

Source: Gemini API deprecations; our capture of 9 Oct 2026, snapshot b3ebdb470ed4.

MaterialGooglelifecycleAPI

Gemini API deprecations

Google released gemini-nano-banana-2.1 on October 6, 2026 as the recommended image replacement, set gemini-3.1-flash-image to shut down March 15, 2027, and transcribe models November 17, 2026.

Changed 6 Oct to 7 Oct

How we know

The wording under “Nano Banana models” (up to 30 words each):

Before: gemini-2.5-flash-image · gemini-2.5-flash-image

After: gemini-nano-banana-2.1 · October 6, 2026 · gemini-3.1-flash-lite-image · June 30, 2026 · gemini-nano-banana-2.1 · March 15, 2027 · gemini-3.1-flash-lite-image · gemini-nano-banana-2.1 · gemini-nano-banana-2.1 · gemini-nano-banana-2.1

Window: last capture of the old text 6 Oct 2026, first capture of the new text 7 Oct 2026 (1 day).

Sections: Audio models; Gemini 3 models; Imagen models; Live API models.

Topics: lifecycle.

Source: Gemini API deprecations; our capture of 7 Oct 2026, snapshot 0698cbcf245b.

MaterialGooglelifecycleBusiness and enterprise

Vertex AI model versions and lifecycle

Google added gemini-nano-banana-2.1 to the shorter availability models table, with a release date of October 6, 2026 and no retirement date announced yet.

Changed 6 Oct to 7 Oct

How we know

The wording under “Models available for shorter availability periods” (up to 30 words each):

Before: nothing (new text).

After: No retirement date announced · gemini-nano-banana-2.1 · October 6, 2026

Window: last capture of the old text 6 Oct 2026, first capture of the new text 7 Oct 2026 (1 day).

Sections: Models available for shorter availability periods.

Topics: lifecycle.

Source: Vertex AI model versions and lifecycle; our capture of 7 Oct 2026, snapshot fdbca74c79eb.

MaterialGooglelifecycleBusiness and enterprise

Vertex AI model versions and lifecycle

Google announced retirement dates for gemini-3.8-flash-cyber, now retiring January 28, 2027, and gemini-3.8-flash, now retiring November 19, 2026; previously neither model had an announced retirement date.

Changed 5 Oct to 6 Oct

How we know

The wording under “Models available for shorter availability periods” (up to 30 words each):

Before: No retirement date announced · No retirement date announced

After: January 28, 2027 · November 19, 2026

Window: last capture of the old text 5 Oct 2026, first capture of the new text 6 Oct 2026 (1 day).

Sections: Models available for shorter availability periods.

Topics: lifecycle.

Source: Vertex AI model versions and lifecycle; our capture of 6 Oct 2026, snapshot 6f1fd887ba4f.

Notable changes

Real wording changes that do not change those obligations. 39 editorial changes (formatting, dates, navigation) are counted but not listed.

NotableAnthropicconsentAPIBusiness and enterpriseConsumer paid

Usage Policy

The policy now explicitly covers app users, API developers, cloud and reseller customers, and end users, and bans unauthorized reselling or proxying of Claude, including through consumer subscriptions.

Changed 8 Oct to 9 Oct

How we know

The wording under “Opening text” (up to 30 words each):

Before: Our Usage Policy (also referred to as our “Acceptable Use Policy” or “AUP”) applies to anyone who can submit inputs to Anthropic’s products and/or services, including via any authorized resellers ...

After: Our Usage Policy (also referred to as our “Acceptable Use Policy” or “AUP”) applies to anyone who submits inputs to Anthropic’s products or services, all of whom we refer to ...

Window: last capture of the old text 8 Oct 2026, first capture of the new text 9 Oct 2026 (1 day).

Sections: Opening text; Additional Use Case Guidelines; Do Not Abuse Our Platform; Do Not Compromise Children’s Safety.

Topics: consent, data sharing, IP and ownership, law and disputes, liability, residency, training.

Source: Usage Policy; our capture of 9 Oct 2026, snapshot 951408a5b007.

NotablexAIdata sharingAPIBusiness and enterprise

Subprocessor list

The subprocessor list now names SpaceXSI instead of SpaceXAI as the company engaging subprocessors under its Data Processing Agreement, and it lists SpaceXSI Affiliates in place of SpaceXAI Affiliates.

Changed 10 Oct to 11 Oct

How we know

The wording under “SpaceXSI Subprocessor List” (up to 30 words each):

Before: This page identifies Subprocessors SpaceXAI has engaged to provide processing activities as detailed in the SpaceXAI Data Processing Agreement. · SpaceXAI Affiliates

After: This page identifies Subprocessors SpaceXSI has engaged to provide processing activities as detailed in the SpaceXSI Data Processing Agreement. · SpaceXSI Affiliates

Window: last capture of the old text 10 Oct 2026, first capture of the new text 11 Oct 2026 (1 day).

Sections: SpaceXSI Subprocessor List.

Topics: data sharing.

Source: Subprocessor list; our capture of 11 Oct 2026, snapshot 1e61bfa4cf0f.

NotableCursorBusiness and enterpriseConsumer paid

Privacy Policy

Cursor now collects payment information both when you seek to access paid products and services and when you initiate payment for products and services, and no longer says "Anysphere" products.

Changed 11 Oct to 12 Oct

How we know

The wording under “1. Personal data we collect” (up to 30 words each):

Before: Payment Information: We collect your payment information if you seek to access any paid Anysphere products and services.

After: Payment Information: We collect your payment information if you seek to access any paid products and services, or initiate payment for products and services.

Window: last capture of the old text 11 Oct 2026, first capture of the new text 12 Oct 2026 (1 day).

Sections: 1. Personal data we collect.

Source: Privacy Policy; our capture of 12 Oct 2026, snapshot 9d2431cda6fe.

NotablexAIConsumer paid

Terms of Service (Consumer)

The terms now call the company SpaceXSI instead of SpaceXAI, including for its APIs, while the contracting entity remains SpaceXAI LLC, a Nevada company at the same Austin address.

Changed 10 Oct to 11 Oct

How we know

The wording under “1. Who We Are” (up to 30 words each):

Before: SpaceXAI is a company working on building artificial intelligence to accelerate human scientific discovery. We are guided by our mission to advance our collective understanding of the universe. As part ...

After: SpaceXSI is a company working on building artificial intelligence to accelerate human scientific discovery. We are guided by our mission to advance our collective understanding of the universe. As part ...

Window: last capture of the old text 10 Oct 2026, first capture of the new text 11 Oct 2026 (1 day).

Sections: 1. Who We Are; Welcome to SpaceXSI.

Source: Terms of Service (Consumer); our capture of 11 Oct 2026, snapshot d73d8962df04.

NotablexAIAPI

API security FAQ (training and retention)

Under ZDR, image generation can now upload results to your own pre-signed URLs instead of only returning base64, and the Console video playground now works using your pre-signed upload URL.

Changed 8 Oct to 9 Oct

How we know

The wording under “Features not supported with ZDR” (up to 30 words each):

Before: Image and video generation can have SpaceXAI host the generated media and return it as a file_id or URL. Under ZDR there is no SpaceXAI-side output storage: no file_id inputs, ...

After: Image and video generation can have SpaceXAI host the generated media and return it as a file_id or URL. Under ZDR there is no SpaceXAI-side output storage: no file_id inputs ...

Window: last capture of the old text 8 Oct 2026, first capture of the new text 9 Oct 2026 (1 day).

Sections: Features not supported with ZDR.

Source: API security FAQ (training and retention); our capture of 9 Oct 2026, snapshot b85b933bdbf8.

NotableMetaAPI

Llama community license

The Community License Agreement section no longer lists Muse Spark 1.3, Muse Spark 1.2, Muse Spark 1.1 or SAM 3.1.

Changed 7 Oct to 8 Oct

How we know

The wording under “Community License Agreement” (up to 30 words each):

Before: Muse Spark 1.3 · Muse Spark 1.2 · Muse Spark 1.1 · SAM 3.1

After: removed.

Window: last capture of the old text 7 Oct 2026, first capture of the new text 8 Oct 2026 (1 day).

Sections: Community License Agreement.

Source: Llama community license; our capture of 8 Oct 2026, snapshot 179c741a8170.

NotableMetaAPI

Llama acceptable use policy

The Acceptable Use Policy page no longer lists Muse Spark 1.3, Muse Spark 1.2, Muse Spark 1.1 or SAM 3.1, and nothing was added in their place.

Changed 7 Oct to 8 Oct

How we know

The wording under “Acceptable Use Policy” (up to 30 words each):

Before: Muse Spark 1.3 · Muse Spark 1.2 · Muse Spark 1.1 · SAM 3.1

After: removed.

Window: last capture of the old text 7 Oct 2026, first capture of the new text 8 Oct 2026 (1 day).

Sections: Acceptable Use Policy.

Source: Llama acceptable use policy; our capture of 8 Oct 2026, snapshot cda8a5deaa65.

NotableGoogleConsumer paid

Gemini Apps Privacy Hub

Google added text saying that, depending on eligibility and settings, Gemini can remember facts, interests and preferences across conversations, and users can ask it to remember details.

Changed 6 Oct to 7 Oct

How we know

The wording under “Keep in mind” (up to 30 words each):

Before: nothing (new text).

After: Depending on eligibility and your settings, Gemini can remember specific facts, interests, and preferences across conversations so you don’t have to repeat info in new chats. You can ask Gemini ...

Window: last capture of the old text 6 Oct 2026, first capture of the new text 7 Oct 2026 (1 day).

Sections: Keep in mind.

Source: Gemini Apps Privacy Hub; our capture of 7 Oct 2026, snapshot fee7e16ebe5c.

NotableGoogle

Generative AI in Google Workspace Privacy Hub

A new answer says files accessible to external collaborators or created outside your domain show an out-of-domain banner in Gemini, and Drive and Gmail grounding is disabled by default.

Changed 6 Oct to 7 Oct

How we know

The wording (up to 30 words each):

Before: nothing (new text).

After: To prevent accidental exposure of organizational data during external collaboration, Gemini incorporates built-in cross-domain guardrails: · Contextual visibility reminder: When a file is accessible to external collaborators (for example, if ...

Window: last capture of the old text 6 Oct 2026, first capture of the new text 7 Oct 2026 (1 day).

Sections: Updates.

Source: Generative AI in Google Workspace Privacy Hub; our capture of 7 Oct 2026, snapshot 65b164d582dd.

NotableAnthropiclifecycleAPIBusiness and enterprise

Model deprecations

The temperature, top_p and top_k deprecation now covers all Claude 4.7 and later models, not just Opus, and deprecated Sonnet models now point to claude-sonnet-5-5 instead of claude-sonnet-4-6.

Changed 9 Oct to 10 Oct

How we know

The wording under “2025-08-13: Claude Sonnet 3.5 models” (up to 30 words each):

Before: claude-sonnet-4-6 · claude-sonnet-4-6

After: claude-sonnet-5-5

Window: last capture of the old text 9 Oct 2026, first capture of the new text 10 Oct 2026 (1 day).

Sections: API parameter deprecations; 2025-08-13: Claude Sonnet 3.5 models; 2025-10-28: Claude Sonnet 3.7 model; 2026-04-14: Claude Sonnet 4 and Claude Opus 4 models.

Topics: lifecycle.

Source: Model deprecations; our capture of 10 Oct 2026, snapshot a374e4d40250.

NotableGooglelifecycleAPI

Gemini API deprecations

Google listed gemini-3.8-flash and gemini-3.6-flash as recommended replacements, saying requests to gemini-3.7-flash and gemini-3.5-flash are automatically routed to them respectively.

Changed 9 Oct to 10 Oct

How we know

The wording under “Gemini 3 models” (up to 30 words each):

Before: nothing (new text).

After: gemini-3.8-flash (requests to gemini-3.7-flash are automatically routed to gemini-3.8-flash) · gemini-3.6-flash (requests to gemini-3.5-flash are automatically routed to gemini-3.6-flash)

Window: last capture of the old text 9 Oct 2026, first capture of the new text 10 Oct 2026 (1 day).

Sections: Gemini 3 models.

Topics: lifecycle.

Source: Gemini API deprecations; our capture of 10 Oct 2026, snapshot 66c3cf046733.

NotableAnthropiclifecycleAPIBusiness and enterprise

Model deprecations

The model status table now lists claude-haiku-5-5 as Active, with no deprecation date (N/A) and a tentative retirement date not sooner than October 7, 2027.

Changed 7 Oct to 8 Oct

How we know

The wording under “Model status” (up to 30 words each):

Before: nothing (new text).

After: Active · N/A · claude-haiku-5-5 · Not sooner than October 7, 2027

Window: last capture of the old text 7 Oct 2026, first capture of the new text 8 Oct 2026 (1 day).

Sections: Model status.

Topics: lifecycle.

Source: Model deprecations; our capture of 8 Oct 2026, snapshot b9a51f310887.

NotableCoherelifecycleAPIBusiness and enterprise

Model deprecations

Cohere only added a note linking to documentation indexes for the v2 and v1 APIs, and no deprecated models, endpoints or replacements changed for customers.

Changed 6 Oct to 7 Oct

How we know

The wording under “Opening text” (up to 30 words each):

Before: nothing (new text).

After: This page is for version v2 API (default). For other versions, use one of these documentation indexes: - v2 API (default): https://docs.cohere.com/v2/llms.txt - v1 API: https://docs.cohere.com/v1/llms.txt

Window: last capture of the old text 6 Oct 2026, first capture of the new text 7 Oct 2026 (1 day).

Sections: Opening text.

Topics: lifecycle.

Source: Model deprecations; our capture of 7 Oct 2026, snapshot 85782bfa57b9.

Found in newly tracked documents

These surfaced this week but may be months old: we only just started tracking the page, or the change can only be dated to more than two weeks. They are not in this week's numbers.

MaterialAWSOlder changeconsentdata sharingAPI

AWS Service Terms, section 50 (machine learning and AI services, including Bedrock)

AWS added Elemental Inference, Continuum (including former Security Agent features), FinOps Agent (Preview) and IoT SiteWise Scenario Discovery as AI Services, and AWS may use and store FinOps Agent AI Content.

Some time in the 304 days, about 10 months before 6 Oct; we began tracking this page on 6 Oct · confirmed by 4 of 4 models

How we know

The wording under “50. AWS Machine Learning and Artificial Intelligence Services” (up to 30 words each):

Before: 50.1. “AI Services” means, collectively, Amazon Bedrock, Amazon CodeGuru Profiler, Amazon CodeGuru Reviewer, Amazon Foundation Models (Amazon Nova and Titan models), Amazon Nova Act, Amazon Nova Forge, Amazon Comprehend, Amazon ...

After: 50.1. “AI Services” means, collectively, Amazon Bedrock, Amazon CodeGuru Profiler, Amazon CodeGuru Reviewer, Amazon Foundation Models (Amazon Nova and Titan models), Amazon Nova Act, Amazon Nova Forge, Amazon Comprehend, Amazon ...

Window: last capture of the old text 6 Dec 2025, first capture of the new text 6 Oct 2026 (304 days, about 10 months).

The old text is an Internet Archive capture; our daily captures of this page began 6 Oct 2026, so the change may be months old and is not counted in this week's numbers.

Sections: 50. AWS Machine Learning and Artificial Intelligence Services.

Topics: consent, data sharing, human review, law and disputes, liability, notice, residency, retention, training.

Graded Material by our published rubric (rules M1) and confirmed by 4 of 4 models reading the changed lines.

Source: AWS Service Terms, section 50 (machine learning and AI services, including Bedrock); our capture of 6 Oct 2026, snapshot 9b46a5beb45e.

MaterialGoogleOlder changedata sharingBusiness and enterprise

Google Cloud Platform subprocessors

Atos's Eviden subprocessors were replaced by Bull entities whose parent is Meteore Holding, a subsidiary of APE, and Google Czech Republic, s.r.o. was added as a Google Group subprocessor.

Some time in the 217 days, about 7 months before 6 Oct; we began tracking this page on 6 Oct · confirmed by 4 of 4 models

How we know

The wording under “Third-Party Subprocessors” (up to 30 words each):

Before: All AI Solutions, Pre-Trained APIs, AI Platform/Vertex AI, Generative AI Services and Agentic AI Services · Atos International, B.V. · Atos International, B.V. · Eviden Canada, Inc. · 3 Church ...

After: N/A · All AI Solutions, Pre-Trained APIs, AI Platform/Gemini Enterprise Agent Platform (formerly Vertex AI), Generative AI Services and Agentic AI Services · N/A · Mexico, Poland, and United States ...

Window: last capture of the old text 3 Mar 2026, first capture of the new text 6 Oct 2026 (217 days, about 7 months).

The old text is an Internet Archive capture; our daily captures of this page began 6 Oct 2026, so the change may be months old and is not counted in this week's numbers.

Sections: Further Information about Google Cloud Platform Subprocessors; Google Group Subprocessors; Third-Party Subprocessors.

Topics: data sharing.

Graded Material by our published rubric (rules M1) and confirmed by 4 of 4 models reading the changed lines.

Source: Google Cloud Platform subprocessors; our capture of 6 Oct 2026, snapshot 2c9ed72d8deb.

MaterialOpenAIOlder changedata sharingAPIBusiness and enterprise

Sub-processor list

OpenAI added a ChatGPT Sites note: listed subprocessors may host, moderate and support Sites web pages, and hosting providers may run safety classifiers and share results with OpenAI.

Some time in the 90 days, about 3 months before 6 Oct; we began tracking this page on 6 Oct · confirmed by 3 of 4 models

How we know

The wording under “Processing for ChatGPT Pro and Plus users:” (up to 30 words each):

Before: nothing (new text).

After: ChatGPT Sites: The OpenAI Subprocessors listed above may provide web hosting, infrastructure, content moderation and support services for ChatGPT Sites created web pages. Web Hosting providers may run security and ...

Window: last capture of the old text 8 Jul 2026, first capture of the new text 6 Oct 2026 (90 days, about 3 months).

The old text is an Internet Archive capture; our daily captures of this page began 6 Oct 2026, so the change may be months old and is not counted in this week's numbers.

Sections: Processing for ChatGPT Pro and Plus users:.

Topics: data sharing.

Graded Material by our published rubric (rules M1) and confirmed by 3 of 4 models reading the changed lines.

Source: Sub-processor list; our capture of 6 Oct 2026, snapshot d172f330b5eb.

Model retirements, next 120 days

8days left20 Oct

Google3 models

gemini-2.5-flash → gemini-3.8-flash or gemini-3.5-flash-lite or gemini-3.1-flash-lite
gemini-2.5-flash-lite → gemini-3.8-flash or gemini-3.1-flash-lite or Gemma 4
gemini-2.5-pro → gemini-3.8-flash or gemini-3.5-flash

How we know

gemini-2.5-flash: October 20, 2026 gemini-3.8-flash or gemini-3.5-flash gemini-2.5-flash (Vertex AI model versions and lifecycle, captured 12 Oct 2026)

gemini-2.5-flash-lite: October 20, 2026 gemini-3.8-flash or gemini-3.5-flash gemini-2.5-flash June 17, 2025 October 20, 2026 gemini-3.8-flash or gemini-3.5-flash-lite or gemini-3.1-flash-lite gemini-2.5-flash-lite (Vertex AI model versions and lifecycle, captured 12 Oct 2026)

gemini-2.5-pro: gemini-2.5-pro June 17, 2025 October 20, 2026 (Vertex AI model versions and lifecycle, captured 12 Oct 2026)

Lead time and agreement are in the full matrix's calendar and its CSV.

10days left22 Oct

Google4 models

gemini-omni-flash-preview new this week → gemini-omni-1.1-flash
veo-3.1-fast-generate-preview → gemini-omni-1.1-flash
veo-3.1-generate-preview → gemini-omni-1.1-flash
veo-3.1-lite-generate-preview → gemini-omni-1.1-flash

How we know

gemini-omni-flash-preview: October 22, 2026 (Gemini API deprecations, captured 12 Oct 2026)

veo-3.1-fast-generate-preview: October 22, 2026 (Gemini API deprecations, captured 12 Oct 2026)

veo-3.1-generate-preview: October 22, 2026 (Gemini API deprecations, captured 12 Oct 2026)

veo-3.1-lite-generate-preview: veo-3.1-lite-generate-preview March 31, 2026 October 22, 2026 (Gemini API deprecations, captured 12 Oct 2026)

Lead time and agreement are in the full matrix's calendar and its CSV.

11days left23 Oct

OpenAI29 models

ft-babbage-002 → gpt-5.6-terra
ft-davinci-002 → gpt-5.6-terra
ft-gpt-3.5-turbo → gpt-5.6-terra
ft-gpt-4 → gpt-5.6-sol
ft-gpt-4.1-nano-2025-04-14 → gpt-5.6-luna
ft-o4-mini-2025-04-16 → gpt-5.6-terra
gpt-3.5-turbo → gpt-5.6-terra
gpt-3.5-turbo-0125 → gpt-5.6-terra
gpt-3.5-turbo-completions → gpt-5.6-terra
gpt-4 → gpt-5.6-sol
gpt-4-0613 → gpt-5.6-sol
gpt-4-0613-completions → gpt-5.6-sol
gpt-4-1106-preview → gpt-5.6-sol
gpt-4-completions → gpt-5.6-sol
gpt-4-turbo → gpt-5.6-sol
gpt-4-turbo-2024-04-09 → gpt-5.6-sol
gpt-4-turbo-completions → gpt-5.6-sol
gpt-4.1-nano → gpt-5.6-luna
gpt-4.1-nano-2025-04-14 → gpt-5.6-luna
gpt-4o-2024-05-13 → gpt-5.6-sol
gpt-image-1 → gpt-image-2.5-sunburst or gpt-image-2.5-flare
o1 → gpt-5.6-sol
o1-2024-12-17 → gpt-5.6-sol
o1-pro → gpt-5.6-sol (reasoning.mode: pro)
o1-pro-2025-03-19 → gpt-5.6-sol (reasoning.mode: pro)
o3-mini → gpt-5.6-sol
o3-mini-2025-01-31 → gpt-5.6-sol
o4-mini → gpt-5.6-terra
o4-mini-2025-04-16 → gpt-5.6-terra
The 6 models starting ft- are fine-tuned models: the replacement has to be fine-tuned again.

How we know

ft-babbage-002: October 23, 2026 (API deprecations, captured 12 Oct 2026)

ft-davinci-002: October 23, 2026 (API deprecations, captured 12 Oct 2026)

ft-gpt-3.5-turbo: October 23, 2026 (API deprecations, captured 12 Oct 2026)

ft-gpt-4: October 23, 2026 (API deprecations, captured 12 Oct 2026)

ft-gpt-4.1-nano-2025-04-14: October 23, 2026 (API deprecations, captured 12 Oct 2026)

ft-o4-mini-2025-04-16: October 23, 2026 (API deprecations, captured 12 Oct 2026)

gpt-3.5-turbo: October 23, 2026 gpt-3.5-turbo (API deprecations, captured 12 Oct 2026)

gpt-3.5-turbo-0125: October 23, 2026 gpt-3.5-turbo-0125 (API deprecations, captured 12 Oct 2026)

gpt-3.5-turbo-completions: October 23, 2026 gpt-3.5-turbo-0125 | gpt-3.5-turbo, gpt-3.5-turbo-completions (API deprecations, captured 12 Oct 2026)

gpt-4: October 23, 2026 gpt-3.5-turbo-0125 | gpt-3.5-turbo, gpt-3.5-turbo-completions gpt-5.6-terra October 23, 2026 gpt-4 (API deprecations, captured 12 Oct 2026)

gpt-4-0613: October 23, 2026 gpt-3.5-turbo-0125 | gpt-3.5-turbo, gpt-3.5-turbo-completions gpt-5.6-terra October 23, 2026 gpt-4-0613 (API deprecations, captured 12 Oct 2026)

gpt-4-0613-completions: October 23, 2026 gpt-3.5-turbo-0125 | gpt-3.5-turbo, gpt-3.5-turbo-completions gpt-5.6-terra October 23, 2026 gpt-4-0613 | gpt-4, gpt-4-0613-completions (API deprecations, captured 12 Oct 2026)

gpt-4-1106-preview: October 23, 2026 gpt-3.5-turbo-0125 | gpt-3.5-turbo, gpt-3.5-turbo-completions gpt-5.6-terra October 23, 2026 gpt-4-0613 | gpt-4, gpt-4-0613-completions, gpt-4-completions gpt-5.6-sol October 23, 2026 gpt-4-1106-preview (API deprecations, captured 12 Oct 2026)

gpt-4-completions: October 23, 2026 gpt-3.5-turbo-0125 | gpt-3.5-turbo, gpt-3.5-turbo-completions gpt-5.6-terra October 23, 2026 gpt-4-0613 | gpt-4, gpt-4-0613-completions, gpt-4-completions (API deprecations, captured 12 Oct 2026)

gpt-4-turbo: October 23, 2026 gpt-3.5-turbo-0125 | gpt-3.5-turbo, gpt-3.5-turbo-completions gpt-5.6-terra October 23, 2026 gpt-4-0613 | gpt-4, gpt-4-0613-completions, gpt-4-completions gpt-5.6-sol October 23, 2026 gpt-4-1106-preview gpt-5.6-sol October 23, 2026 gpt-4-turbo (API deprecations, captured 12 Oct 2026)

gpt-4-turbo-2024-04-09: October 23, 2026 gpt-3.5-turbo-0125 | gpt-3.5-turbo, gpt-3.5-turbo-completions gpt-5.6-terra October 23, 2026 gpt-4-0613 | gpt-4, gpt-4-0613-completions, gpt-4-completions gpt-5.6-sol October 23, 2026 gpt-4-1106-preview gpt-5.6-sol October 23, 2026 gpt-4-turbo | gpt-4-turbo-2024-04-09 (API deprecations, captured 12 Oct 2026)

gpt-4-turbo-completions: October 23, 2026 gpt-3.5-turbo-0125 | gpt-3.5-turbo, gpt-3.5-turbo-completions gpt-5.6-terra October 23, 2026 gpt-4-0613 | gpt-4, gpt-4-0613-completions, gpt-4-completions gpt-5.6-sol October 23, 2026 gpt-4-1106-preview gpt-5.6-sol October 23, 2026 gpt-4-turbo | gpt-4-turbo-2024-04-09, gpt-4-turbo-completions (API deprecations, captured 12 Oct 2026)

gpt-4.1-nano: October 23, 2026 (API deprecations, captured 12 Oct 2026)

gpt-4.1-nano-2025-04-14: October 23, 2026 (API deprecations, captured 12 Oct 2026)

gpt-4o-2024-05-13: October 23, 2026 (API deprecations, captured 12 Oct 2026)

gpt-image-1: October 23, 2026 (API deprecations, captured 12 Oct 2026)

o1: October 23, 2026 (API deprecations, captured 12 Oct 2026)

o1-2024-12-17: October 23, 2026 (API deprecations, captured 12 Oct 2026)

o1-pro: October 23, 2026 (API deprecations, captured 12 Oct 2026)

o1-pro-2025-03-19: October 23, 2026 (API deprecations, captured 12 Oct 2026)

o3-mini: October 23, 2026 (API deprecations, captured 12 Oct 2026)

o3-mini-2025-01-31: October 23, 2026 (API deprecations, captured 12 Oct 2026)

o4-mini: October 23, 2026 (API deprecations, captured 12 Oct 2026)

o4-mini-2025-04-16: October 23, 2026 (API deprecations, captured 12 Oct 2026)

Lead time and agreement are in the full matrix's calendar and its CSV.

Later: 32 models from 17 Nov 2026 to 28 Jan 2027

36days left17 Nov

Google5 models

gemini-2.5-flash-native-audio-preview-12-2025 new this week → gemini-3.8-live
gemini-2.5-flash-preview-tts new this week → gemini-3.8-flash-tts or gemini-3.8-flash-lite-tts
gemini-2.5-pro-preview-tts new this week → gemini-3.8-flash-tts or gemini-3.8-flash-lite-tts
gemini-3.1-flash-live-preview new this week → gemini-3.8-live
gemini-3.1-flash-tts-preview new this week → gemini-3.8-flash-tts or gemini-3.8-flash-lite-tts

How we know

gemini-2.5-flash-native-audio-preview-12-2025: November 17, 2026 gemini-3.8-live gemini-2.5-flash-native-audio-preview-12-2025 (Gemini API deprecations, captured 12 Oct 2026)

gemini-2.5-flash-preview-tts: November 17, 2026 (Gemini API deprecations, captured 12 Oct 2026)

gemini-2.5-pro-preview-tts: November 17, 2026 (Gemini API deprecations, captured 12 Oct 2026)

gemini-3.1-flash-live-preview: gemini-3.1-flash-live-preview March 11, 2026 November 17, 2026 (Gemini API deprecations, captured 12 Oct 2026)

gemini-3.1-flash-tts-preview: gemini-3.1-flash-tts-preview February 26, 2026 November 17, 2026 (Gemini API deprecations, captured 12 Oct 2026)

Lead time and agreement are in the full matrix's calendar and its CSV.

38days left19 Nov

Google1 model

gemini-3.6-flash new this week → gemini-3.8-flash

How we know

gemini-3.6-flash: gemini-3.6-flash July 21, 2026 November 19, 2026 (Vertex AI model versions and lifecycle, captured 12 Oct 2026)

Lead time and agreement are in the full matrix's calendar and its CSV.

49days left30 Nov

Anthropic1 model

claude-sonnet-4-5-20250929 → claude-sonnet-5-5

How we know

claude-sonnet-4-5-20250929: claude-sonnet-4-5-20250929 Deprecated September 30, 2026 November 30, 2026 (Model deprecations, captured 12 Oct 2026)

Lead time and agreement are in the full matrix's calendar and its CSV.

50days left1 Dec

OpenAI3 models

chatgpt-image-latest → gpt-image-2.5-sunburst or gpt-image-2.5-flare
gpt-image-1-mini → gpt-image-2.5-sunburst or gpt-image-2.5-flare
gpt-image-1.5 → gpt-image-2.5-sunburst or gpt-image-2.5-flare

How we know

chatgpt-image-latest: On June 2, 2026, we notified developers using older GPT Image models of their deprecation and removal from the API on December 1, 2026. (API deprecations, captured 12 Oct 2026)

gpt-image-1-mini: On June 2, 2026, we notified developers using older GPT Image models of their deprecation and removal from the API on December 1, 2026. (API deprecations, captured 12 Oct 2026)

gpt-image-1.5: On June 2, 2026, we notified developers using older GPT Image models of their deprecation and removal from the API on December 1, 2026. (API deprecations, captured 12 Oct 2026)

Lead time and agreement are in the full matrix's calendar and its CSV.

58days left9 Dec

Microsoft1 model

gpt-4o → gpt-5.6-sol

How we know

gpt-4o: When gpt-4o version 2024-05-13 retires on 2026-12-09, the service automatically upgrades it to gpt-5.6-sol on the Standard SKU in each region where that version is currently available. If gpt-5.6-sol doesn't yet have a Standard presence in one of those regions, the upgrade process adds it there. Check the Model Retirement Schedule for the current retirement date and replacement model before this upgrade occurs, since these details are subject to change. (Azure OpenAI model deprecations & retirements, captured 12 Oct 2026)

Lead time and agreement are in the full matrix's calendar and its CSV.

60days left11 Dec

OpenAI6 models

gpt-5-2025-08-07 → gpt-5.6-sol
gpt-5-mini-2025-08-07 → gpt-5.6-terra
gpt-5-nano-2025-08-07 → gpt-5.6-luna
gpt-5-pro-2025-10-06 → gpt-5.6-sol (reasoning.mode: pro)
o3-2025-04-16 → gpt-5.6-sol
o3-pro-2025-06-10 → gpt-5.6-sol (reasoning.mode: pro)

How we know

gpt-5-2025-08-07: On June 11, 2026, we notified developers using older GPT-5 and o3 model snapshots of their deprecation and removal from the API on December 11, 2026. (API deprecations, captured 12 Oct 2026)

gpt-5-mini-2025-08-07: On June 11, 2026, we notified developers using older GPT-5 and o3 model snapshots of their deprecation and removal from the API on December 11, 2026. (API deprecations, captured 12 Oct 2026)

gpt-5-nano-2025-08-07: On June 11, 2026, we notified developers using older GPT-5 and o3 model snapshots of their deprecation and removal from the API on December 11, 2026. (API deprecations, captured 12 Oct 2026)

gpt-5-pro-2025-10-06: On June 11, 2026, we notified developers using older GPT-5 and o3 model snapshots of their deprecation and removal from the API on December 11, 2026. (API deprecations, captured 12 Oct 2026)

o3-2025-04-16: On June 11, 2026, we notified developers using older GPT-5 and o3 model snapshots of their deprecation and removal from the API on December 11, 2026. (API deprecations, captured 12 Oct 2026)

o3-pro-2025-06-10: On June 11, 2026, we notified developers using older GPT-5 and o3 model snapshots of their deprecation and removal from the API on December 11, 2026. (API deprecations, captured 12 Oct 2026)

Lead time and agreement are in the full matrix's calendar and its CSV.

62days left13 Dec

Google1 model

gemini-live-2.5-flash-native-audio (no replacement named)

How we know

gemini-live-2.5-flash-native-audio: gemini-live-2.5-flash-native-audio December 12, 2025 December 13, 2026 (Vertex AI model versions and lifecycle, captured 12 Oct 2026)

Lead time and agreement are in the full matrix's calendar and its CSV.

86days left6 Jan

OpenAI4 models

gpt-4o-mini-tts-2025-03-20 → gpt-realtime-2.1-mini
gpt-4o-mini-tts-2025-12-15 → gpt-realtime-2.1-mini
tts-1 → gpt-realtime-2.1-mini
tts-1-hd → gpt-realtime-2.1-mini

How we know

gpt-4o-mini-tts-2025-03-20: The following text-to-speech models are deprecated and will be removed from the API on January 6, 2027, with at least three months’ notice. Migrate to gpt-realtime-2.1-mini before the shutdown date. See the Realtime API guide to plan your migration. (API deprecations, captured 12 Oct 2026)

gpt-4o-mini-tts-2025-12-15: The following text-to-speech models are deprecated and will be removed from the API on January 6, 2027, with at least three months’ notice. Migrate to gpt-realtime-2.1-mini before the shutdown date. See the Realtime API guide to plan your migration. (API deprecations, captured 12 Oct 2026)

tts-1: The following text-to-speech models are deprecated and will be removed from the API on January 6, 2027, with at least three months’ notice. Migrate to gpt-realtime-2.1-mini before the shutdown date. See the Realtime API guide to plan your migration. (API deprecations, captured 12 Oct 2026)

tts-1-hd: The following text-to-speech models are deprecated and will be removed from the API on January 6, 2027, with at least three months’ notice. Migrate to gpt-realtime-2.1-mini before the shutdown date. See the Realtime API guide to plan your migration. (API deprecations, captured 12 Oct 2026)

Lead time and agreement are in the full matrix's calendar and its CSV.

100days left20 Jan

OpenAI9 models

gpt-4o-audio → gpt-audio-1.5
gpt-4o-mini-audio → gpt-audio-1.5
gpt-4o-mini-realtime → gpt-realtime-2.1-mini
gpt-4o-mini-transcribe-2025-03-20 → gpt-4o-mini-transcribe-2025-12-15
gpt-4o-realtime → gpt-realtime-2.1
gpt-audio → gpt-audio-1.5
gpt-audio-mini → gpt-audio-1.5
gpt-realtime → gpt-realtime-2.1
gpt-realtime-mini → gpt-realtime-2.1-mini

How we know

gpt-4o-audio: On July 20, 2026, we notified developers using legacy audio, realtime, and transcription model families and snapshots of their deprecation and removal from the API on January 20, 2027. (API deprecations, captured 12 Oct 2026)

gpt-4o-mini-audio: On July 20, 2026, we notified developers using legacy audio, realtime, and transcription model families and snapshots of their deprecation and removal from the API on January 20, 2027. (API deprecations, captured 12 Oct 2026)

gpt-4o-mini-realtime: On July 20, 2026, we notified developers using legacy audio, realtime, and transcription model families and snapshots of their deprecation and removal from the API on January 20, 2027. (API deprecations, captured 12 Oct 2026)

gpt-4o-mini-transcribe-2025-03-20: otified developers using legacy audio, realtime, and transcription model families and snapshots of their deprecation and removal from the API on January 20, 2027. (API deprecations, captured 12 Oct 2026)

gpt-4o-realtime: On July 20, 2026, we notified developers using legacy audio, realtime, and transcription model families and snapshots of their deprecation and removal from the API on January 20, 2027. (API deprecations, captured 12 Oct 2026)

gpt-audio: On July 20, 2026, we notified developers using legacy audio, realtime, and transcription model families and snapshots of their deprecation and removal from the API on January 20, 2027. (API deprecations, captured 12 Oct 2026)

gpt-audio-mini: On July 20, 2026, we notified developers using legacy audio, realtime, and transcription model families and snapshots of their deprecation and removal from the API on January 20, 2027. (API deprecations, captured 12 Oct 2026)

gpt-realtime: On July 20, 2026, we notified developers using legacy audio, realtime, and transcription model families and snapshots of their deprecation and removal from the API on January 20, 2027. (API deprecations, captured 12 Oct 2026)

gpt-realtime-mini: On July 20, 2026, we notified developers using legacy audio, realtime, and transcription model families and snapshots of their deprecation and removal from the API on January 20, 2027. (API deprecations, captured 12 Oct 2026)

Lead time and agreement are in the full matrix's calendar and its CSV.

108days left28 Jan

Google1 model

gemini-3.7-flash new this week → gemini-3.8-flash

How we know

gemini-3.7-flash: gemini-3.7-flash August 13, 2026 January 28, 2027 (Vertex AI model versions and lifecycle, captured 12 Oct 2026)

Lead time and agreement are in the full matrix's calendar and its CSV.

1 date in review

Some models listed these dates but fewer than three agreed, so they are not in the calendar or its counts. Check the vendor's page.

deep-research-pro-preview-12-2025, Google, 23 Oct 2026 (11 days left): deep-research-pro-preview-12-2025 December 11, 2025 October 23, 2026 (2 of 4 models listed it)

Each date was agreed by at least three of four models and found next to the model name in the captured page.

Matrix changes since the last edition

AnthropicClaude Team and Enterprise

Notice of model retirement

~~60 days~~ → Not stated

How we know

3 of 4 models agreed after a change to Anthropic Model deprecations.

OpenAIOpenAI API

Data residency

~~AU, CA, CH, EEA, IN, JP, KR, SG, UK, US~~ → AE, AU, CA, CH, EEA, IN, JP, KR, SG, UK, US

How we know

Settled by a reviewer: Same call as on 2026-10-05, re-made on the current text: the Support by region table still lists United States, Europe (EEA + Switzerland), Australia, Canada, Japan, India, Singapore, South Korea, United Kingdom and United Arab Emirates. The seats split only on adding EU, which the EEA already includes.

When data residency is enabled on your account, you can set a region for new projects you create in your account from the available regions listed below. (Data controls in the OpenAI platform, How does data residency work?, captured 12 Oct 2026)

GoogleVertex AI (Google Cloud)

Default retention

~~Not stated~~ → 1 day

How we know

3 of 4 models agreed on a second read, after a model failed to answer the first.

This data is stored only in-memory (not at-rest), is isolated at the project level, and has a 24-hour TTL. (Vertex AI (Gemini Enterprise Agent Platform) data governance and zero data retention, In-memory data caching, captured 12 Oct 2026)

CohereCohere enterprise deployments

IP indemnity

~~Conditional~~ → Yes

How we know

3 of 4 models agreed after a change to Cohere Enterprise Data Commitments; Cohere Commercial SaaS Agreement (paid API).

against any and all Losses arising from Claims by a third party alleging that any Output infringes, violates or misappropriates any copyright rights of a third party (Commercial SaaS Agreement (paid API), Cohere Copyright Assurance, captured 12 Oct 2026)

Corrections to the 5 Oct 2026 edition

2 values we printed last time were wrong. None of the documents behind them changed, so these are our corrections, not vendor changes, and they are not counted above.

Anthropic, Claude API, trains on your data by default: printed No, correct value Opt-in. That edition's value rested on only 2 of 3 models; at least 3 must agree. Source: unless the customer has chosen to provide their data to us for model improvement (for example, the Developer Partner Program). (Claude Code data usage)
Anthropic, Claude API, abuse-monitoring retention: printed Not stated, correct value 30 days. That edition's value rested on only 2 of 3 models; at least 3 must agree. Source: For Anthropic API users, we automatically delete inputs and outputs on our backend within 30 days of receipt or generation (Privacy Center: how long do you store my organization's data)

Coverage this week

We capture 59 documents from 13 vendors every day. 3 documents missed a capture this week (a change on a missed day shows up in the next capture, dated to a longer window); 9 documents were added. Degraded sources: none.

Details

Vendors: AWS, Anthropic, Cohere, Cursor, DeepSeek, GitHub Copilot, Google, Meta, Microsoft, Mistral AI, OpenAI, Perplexity, xAI.

Added since the 5 Oct 2026 edition: Anthropic: Data Processing Addendum; AWS: AWS Service Terms, section 50 (machine learning and AI services, including Bedrock); AWS: AWS sub-processors; Cohere: Commercial SaaS Agreement (paid API); GitHub Copilot: GitHub subprocessors; Google: Google Cloud Platform subprocessors; OpenAI: Data Processing Addendum; OpenAI: Sub-processor list; xAI: Subprocessor list. Their history before that comes from the Internet Archive.

Perplexity: API Terms of Service, no page capture on 10 Oct 2026, 11 Oct 2026.

Perplexity: Enterprise Terms and Conditions, no page capture on 10 Oct 2026, 11 Oct 2026.

Perplexity: Terms of Service (consumer, including Pro), no page capture on 10 Oct 2026, 11 Oct 2026.

The full matrix

Every vendor and plan, every cell cited: document, section, quote, capture date and snapshot hash.

Subscribers get the full matrix with every edition: PDF, matrix CSV and changes CSV. See the free 3-vendor sample or the prices.