Change feed

The document rebrands xAI as "SpaceXAI" throughout and updates the effective date. Substantively, the competitive-training restriction is broadened to prohibit assisting others in developing competing models or products/services (directly or indirectly), and the automated-access restriction is narrowed to cover only unauthorized automated or non-human means.

AI-generated description; verify against the source before relying on it.

The Trust Center added a new "Subprocessor Update" entry announcing three new sub-processors-Parallel AI (web search), and QumulusAI and Argentum AI (cloud infrastructure)-while retaining the prior ISO 27001 certification announcement lower on the page.

AI-generated description; verify against the source before relying on it.

The Trust Center page was updated to announce Baseten's new ISO/IEC 27001 certification, replacing a prior SOC 2 Type 2 announcement, and removed the SIG Lite self-assessment reference.

AI-generated description; verify against the source before relying on it.

The document's effective date was updated to August 7th 2026, and an Acceptable Use Policy section was added/introduced, outlining prohibited uses of the Services by Customer and End Users.

AI-generated description; verify against the source before relying on it.

The status of claude-opus-4-1-20250805 was updated from "Deprecated" to "Retired" and a note was added confirming the model was retired on August 5, 2026.

AI-generated description; verify against the source before relying on it.

The document updates the model deprecation table, replacing legacy model version identifiers (e.g., gpt-5.4-mini, gpt-5.5, gpt-5.4-nano) with newer successor models (gpt-5.6-terra, gpt-5.6-sol, gpt-5.6-luna) as recommended replacements, and removes an entry for gpt-4o-mini-tts deprecation dates and versions.

AI-generated description; verify against the source before relying on it.

The update replaces the prior announcement about Baseten's SOC 2 Type 2 report availability with a broader compliance notice referencing both the SOC 2 report and a new independent third-party HIPAA assessment report, both now available in the Trust Center for customer/partner request.

AI-generated description; verify against the source before relying on it.

The document's date stamp was updated, and Section 5 was restructured with numbered subsections (5.1, 5.2, 5.3); a new Section 5.3 was added establishing account responsibility rules, holding the Enterprise User accountable for all acts under its account and specifying that operations by authorized persons are binding on the User.

AI-generated description; verify against the source before relying on it.

The document date was updated and the payment section was renumbered into subsections (5.1, 5.2, 5.3), with a new clause added establishing that the Enterprise User is responsible for all acts and operations conducted under its Account, including actions by authorized persons.

AI-generated description; verify against the source before relying on it.

The sub-processor list was updated: entries for CoreWeave, Cloudflare, Wiz, and CrowdStrike were removed and re-added with revised categorization and links, and a new sub-processor, Merge API Inc., was added to support connector functionality on Vibe.

AI-generated description; verify against the source before relying on it.

The sub-processor list page was updated to reflect an additional control/entry (from 2 to 3 more controls shown), indicating a change to the list of sub-processors disclosed.

AI-generated description; verify against the source before relying on it.

The document's timestamp was updated and the subprocessor list count increased from 13 to 14 additional entries, indicating a new subprocessor was added.

AI-generated description; verify against the source before relying on it.

The clause broadened its scope from applying specifically to Azure OpenAI in Microsoft Foundry Models to applying to any covered service, and updated the reference for required mitigation documentation from the prior Azure OpenAI customer-copyright-commitment URL to a new Customer Copyright Commitment Required Mitigations page.

AI-generated description; verify against the source before relying on it.

The listed document reference was changed from the "Secure AI Frontier Model Framework" description to an "EU AI Act FAQs" document covering questions about the GPAI Code of Practice, and the count of additional controls shown was reduced from 3 to 2.

AI-generated description; verify against the source before relying on it.

The document's timestamp was updated and the counts of additional listed organizational and internal security controls increased, indicating new security control entries were added to the subprocessor listing.

AI-generated description; verify against the source before relying on it.

The deprecations page was updated to add new deprecation notices (e.g., codex-mini-latest snapshot deprecation announced November 17, 2025 with removal February 12, 2026, and chatgpt-4o-latest snapshot removal moved to February 17, 2026 with gpt-5.1-chat-latest as replacement), and to move several previously upcoming shutdowns (legacy GPT model snapshots, gpt-3.5-turbo-instruct/babbage-002/davinci-002/gpt-3.5-turbo-1106, and the Assistants API deprecation) into completed/past status with updated shutdown dates such as March 26, 2026. Some entries and their recommended replacement models were removed or reorganized as they transitioned from upcoming to past deprecations.

AI-generated description; verify against the source before relying on it.

The document's timestamp was updated and the list of organizational security controls was expanded from one additional control to two.

AI-generated description; verify against the source before relying on it.

A hyperlink reference was updated from the general Google data center locations page to a specific Google Cloud data residency terms page for SecOps.

AI-generated description; verify against the source before relying on it.

The sub-processor list count changed, reflecting the removal of one sub-processor entry from the Trust Center listing.

AI-generated description; verify against the source before relying on it.

The update adds several new security-related controls to Crusoe's subprocessor/trust page, including additional infrastructure security controls, an employee Code of Conduct acknowledgment requirement, vulnerability and system monitoring procedures, and continuity/disaster recovery plans, alongside expanded internal security procedure counts.

AI-generated description; verify against the source before relying on it.

The AWS IoT Services list was updated (removing AWS IoT 1-Click and AWS IoT Events/Analytics, adding AWS IoT SiteWise Scenario Discovery), and a new restriction (39.14) was added prohibiting use of AWS IoT SiteWise Scenario Discovery for real-time vehicle control, operational command transmission, direct vehicle integration, or regulatory submissions, with a corresponding clarification that curated data also cannot substitute for human monitoring. Minor edits also appeared in AI Services definitions and Chime PSTN invoicing language, and the document date was updated.

AI-generated description; verify against the source before relying on it.

The sub-processor list was updated to add new entities (CoreWeave Inc. for inference, Mistral Compute for cloud infrastructure, Ory Corp. for authentication) and expand data processing locations to include Belgium alongside existing regions, while removing Get Lago Corp. as a billing sub-processor. The total count of listed sub-processors increased from 14 to 25.

AI-generated description; verify against the source before relying on it.

The document renames the product "Gemini Enterprise for Customer Experience" to "CX Agent Studio" throughout web-grounding related clauses, and adds a new provision addressing Canada Controlled Goods, allowing use of Services with software/technical data subject to Canada's Defence Production Act and Controlled Goods Regulations when using specified Assured Workloads Services. The last modified date was also updated.

AI-generated description; verify against the source before relying on it.

A sub-processor or control listing count was incremented, indicating an additional sub-processor/control was added to the Trust Center page.

AI-generated description; verify against the source before relying on it.

The sub-processor list was updated to add AWS Config and AWS FinOps Agent (with Canada/USA locations), and a new footnote clarifies AWS Config's third-party recorder use alongside the existing CloudWatch AI/ML footnote; the Last Updated date was also revised.

AI-generated description; verify against the source before relying on it.

The sub-processors list control count was updated from 4 additional controls to 3, indicating a change in the number of listed sub-processors or associated controls.

AI-generated description; verify against the source before relying on it.

The subprocessors list was updated to add one additional subprocessor entry, along with a routine timestamp update reflecting the edit.

AI-generated description; verify against the source before relying on it.

The policy renames the product "Le Chat" to "Vibe" throughout, broadens the definition of data covered (now including data accessed via connected integrations/services, not just data provided), and updates the description of data usage for product improvement to reference aggregated or anonymous datasets rather than solely aggregated and anonymous statistics. Notably, the prior explicit statement that Input/Output from Le Chat Enterprise or paid API use is not used for AI model training, and that connected third-party services are not used for model training, has been removed.

AI-generated description; verify against the source before relying on it.

The DPA simplified the opt-out language for AI training use, removing the distinction between products opted-out by default versus requiring explicit opt-in, and now simply states training occurs unless Customer has opted-out. Minor grammatical corrections were also made elsewhere (e.g., 'authorized' to 'authorizes', 'the Mistral AI' to 'Mistral AI') without altering the substance of the data transfer or audit provisions.

AI-generated description; verify against the source before relying on it.

The document's timestamp was refreshed and the subprocessor list was updated to add a new security control item ('Visitor procedures enforced') along with an increase in the count of additional internal security procedures listed from 6 to 8.

AI-generated description; verify against the source before relying on it.

The sub-processors list count indicator was updated from '3 more controls' to '4 more controls', indicating a new sub-processor was added to the list.

AI-generated description; verify against the source before relying on it.

The subprocessors page was updated to add a new listed subprocessor or security control entry, specifically 'Asset disposal procedures utilized,' and the count of additional internal security procedures/controls increased from 5 to 6.

AI-generated description; verify against the source before relying on it.

The document's date stamp was updated and a new entry for Claude Sonnet 5 was added to the Transparency Hub, including a summary table describing its capabilities, benchmarked performance (referencing the system card's Section 8), and acceptable use policy references. A prior reference to the Claude Fable 5 & Claude Mythos 5 system card was removed.

AI-generated description; verify against the source before relying on it.

The document was updated to add a new model entry for Claude Sonnet 5, including its model description, benchmarked capabilities reference, and acceptable use policy summary, alongside a date bump. A prior reference to a 'Claude Fable 5 & Claude Mythos 5 system card' link was removed.

AI-generated description; verify against the source before relying on it.

The deprecation status of Claude Mythos Preview was updated from a scheduled future retirement date (July 21, 2026) to now being marked as already deprecated. A new model entry, claude-opus-5, was added to the deprecation table listing it as Active with no scheduled deprecation date, and not sooner than July 24, 2027 for retirement. Minor formatting changes were also made to table column widths.

AI-generated description; verify against the source before relying on it.

Baseten updated its Trust Center subprocessor list, removing Hubspot, Stripe, Clazar, Salesforce, and Slack because they are no longer used or do not process personal data. The page also reflects reorganized compliance content (e.g., SIG Lite Questionnaire and CSA STAR Level 1 announcements) moved or restated within the Trust Center.

AI-generated description; verify against the source before relying on it.

The trust center page removed the 'Internal Control Monitoring' item and adjusted the controls list, changing the count from 11 to 10 additional controls while adding a link labeled 'Internal Control Policy'.

AI-generated description; verify against the source before relying on it.

A reference to the "Shared Assessments - SIG Lite" item was removed from the sub-processors/Trust Center listing, with no replacement text provided.

AI-generated description; verify against the source before relying on it.

The document's effective date was updated to July 23rd 2026 and new introductory language was added addressing OpenRouter LLC customers, describing the Crusoe Managed Inference Service offering access to AI models powered by GPUs.

AI-generated description; verify against the source before relying on it.

The document was updated with a new effective date and added a section introducing Terms of Service specifically for OpenRouter Customers, describing access to Crusoe's Managed Inference Service for AI models via OpenRouter LLC.

AI-generated description; verify against the source before relying on it.

Added a new reference link to the ATOM repository and a MiniMax-M3 MXFP4/MXFP8 usage guide for running the model, expanding the documentation resources listed in the model card.

AI-generated description; verify against the source before relying on it.

The document updates the model deprecation and replacement mapping, redirecting older models (gpt-5.5, gpt-5.5-pro, gpt-5.4-mini, gpt-5.4-nano) to new successor models (gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna), and reformats deprecation dates from ISO format to a written month/day/year format.

AI-generated description; verify against the source before relying on it.

The DPA was restructured with a new introductory framing that identifies Runpod as Processor and Customer as Controller, and the definitions section was substantially reworded: several defined terms were revised or removed (e.g., 'Applicable Data Protection Laws' replaced by broader 'Data Protection Laws', GDPR-specific definition removed in favor of a simpler EEA definition), and new terms were added (Data Protection Laws, FADP, FDPIC, ICO, Privacy Statement). Minor wording tweaks were made to Data Subject, Data Subject Request, Personal Data Breach, and Personnel definitions to generalize scope beyond Runpod-specific language.

AI-generated description; verify against the source before relying on it.

The Trust Center sub-processors page updated a control count from '4 more controls' to '3 more controls', indicating a change in the number of listed sub-processors or associated controls.

AI-generated description; verify against the source before relying on it.

Added a new deprecation notice announcing that legacy audio, realtime, and transcription model families and snapshots will be removed from the API on January 20, 2027, with a table listing shutdown dates, affected models, and recommended replacements.

AI-generated description; verify against the source before relying on it.

The document's timestamp was refreshed and the list of subprocessors was updated, reducing the count of additional internal security/procedure entries shown from 11 to 10.

AI-generated description; verify against the source before relying on it.

The Trust Center page updated its control listings, removing the "Risk Register" control and adding a "Vendor Risk Management Policy" control, with corresponding changes to the displayed count of additional controls.

AI-generated description; verify against the source before relying on it.

The document was renamed from "OpenAI Business Terms" to the "OpenAI Services Agreement," with an updated effective date and a substantially restructured/expanded definitions section (including new terms like Dispute, Discloser, Recipient, Governing Laws, IP Rights, Initial Term, Renewal Term, Protected Health Information). The scope clause was revised to clarify which OpenAI products (APIs, ChatGPT Enterprise, ChatGPT Business, ChatGPT for Clinicians) are covered, and provisioning/purchasing mechanics for Authorized Purchasers were added. Governing law language was moved into a new defined "Governing Laws" term tied to Customer location, replacing the prior fixed jurisdiction/choice-of-law clause.

AI-generated description; verify against the source before relying on it.

The document renamed 'reCAPTCHA Enterprise' to 'Google Cloud Fraud Defense' throughout, and added a new section on Generative AI Safety and Abuse allowing Google to use automated tools to detect abuse of Generative AI Services and take action notwithstanding the standard prompt/output handling terms if AUP or Prohibited Use Policy violations are detected. The last modified date was also updated.

AI-generated description; verify against the source before relying on it.

The EC2 Capacity Blocks for ML section was restructured into a broader "EC2 Capacity Reservations" section, adding a new subsection on On-Demand Capacity Reservations that prohibits resale and allows AWS to cancel purchases or terminate instances if resale is suspected, while retaining nonrefundable/noncancelable terms for Capacity Blocks.

AI-generated description; verify against the source before relying on it.