Can My Agent Uselookup · compare · cite

Aug 28, 2026 · Catalog foundation is up

Product feature · security-privacy

Training data controlsCommon product term

Control or document model-training and service-improvement use of product content.

Terminology basis: Common product term. xAI — Grok Bot approvals, security, and privacy

Training data controls: 3 supported, 4 partial, 0 unsupported, 24 unreviewed across 31 cataloged products.

Markdown · JSON

Explore this familyMore in Security and privacy7 capabilities

Current evidence by product

Can my agent use Training data controls?

Read across for the answer. 7 of 31 current product columns have reviewed evidence; unreviewed does not mean unsupported.

Known issues (0)Report an issue
  • Supported3
  • Partial4
  • Unsupported0
  • Unknown24
  • Not applicable0

Web

9 products

Desktop

13 products

CLI

9 products

Unknown means no public evidence has been reviewed for that product and capability. It does not mean unsupported.

How statuses are assigned

Definition and scope

What this capability means

This row separates several uses that are often compressed into “not used for training”: foundation-model training, service improvement, evaluation, personalization, abuse prevention, safety review, human review, and content a user intentionally submits as feedback.

Evidence should identify defaults by account and plan, opt-in or opt-out mechanics, administrator enforcement, whether connectors and tool output follow the same rule, policy-change notice, and whether third-party model or tool providers receive content under different terms.

Traceable compatibility

Assertion ledger

Documentation evidence only. No runtime conformance test is implied.

Grok Botdesktop · current
Partial
Target
2026-08-28 Grok Bot desktop documentation observation · hosted-observation
Environment
hosted-default
Observed
2026-08-28
  • policytraining opt-out follows Cursor account and privacy settings, and an organization's privacy mode governs team members
  • runtimethe reviewed page does not fully distinguish foundation-model training, evaluation, safety review, feedback, human review, and third-party subprocessors
Evidence
ChatGPTweb · current
Partial
Target
2026-08-28 ChatGPT Work documentation observation · hosted-observation
Environment
enterprise-managed
Observed
2026-08-28
  • planChatGPT Business and Enterprise business data is not used for training by default
  • policythis cell does not establish the default, opt-out mechanics, feedback use, or human-review policy for personal plans
Evidence
ChatGPTdesktop · current
Partial
Target
2026-08-28 ChatGPT Work documentation observation · hosted-observation
Environment
enterprise-managed
Observed
2026-08-28
  • planChatGPT Business and Enterprise business data is not used for training by default
  • policythis cell does not establish the default, opt-out mechanics, feedback use, or human-review policy for personal plans
Evidence
Codex CLIcli · current
Partial
Target
2026-08-28 Codex pricing documentation observation · hosted-observation
Environment
enterprise-managed
Observed
2026-08-28
  • planChatGPT Business and Enterprise business data is not used for training by default when Codex is used through the workspace plan
  • authAPI-key authentication follows API data controls and is not established by this ChatGPT-plan cell
Evidence
Claude CLIcli · current
Supported
Target
2026-08-28 Claude Code documentation observation · hosted-observation
Environment
local-default
Observed
2026-08-28
  • planFree, Pro, and Max users can choose whether their data may improve future Claude models; Team, Enterprise, API, third-party platform, and government data is not used to train generative models under commercial terms unless affirmatively provided
  • policy/feedback, /bug, /share, optional transcript surveys, and customer opt-in programs have separate documented retention and improvement rules
Evidence
Geminiweb · current
Supported
Target
2026-08-28 Gemini Apps privacy documentation observation · hosted-observation
Environment
hosted-default
Observed
2026-08-28
  • auththis cell applies to signed-in personal accounts; work and school accounts follow administrator-controlled Workspace terms
  • policywith Keep Activity off, future chats do not train Google's AI models unless feedback is submitted; service delivery and safety processing can still involve human review
  • policyaudio and Gemini Live video or screenshare improvement is controlled by a separate setting and is off by default
Evidence
Cursordesktop · current
Supported
Target
2026-08-28 Cursor privacy documentation observation · hosted-observation
Environment
local-default
Observed
2026-08-28
  • policyPrivacy Mode prevents Customer Data from training Cursor models and uses ZDR agreements with model providers; non-ZDR models require designation or administrator opt-in
  • policysafety classifiers and triggered abuse investigations are exceptions; turning Privacy Mode off permits broader storage, improvement, and model-training use
Evidence
  1. 1. Evidence checked 2026-08-28: xAI documents that Grok Bot training opt-out follows the applicable Cursor account and privacy settings, with organization privacy mode governing while a member is on a team. The reviewed page does not fully separate training, evaluation, safety review, feedback, or every subprocessor's use.
  2. 2. Evidence checked 2026-08-28: OpenAI states that ChatGPT Business and Enterprise do not train on business data by default across ChatGPT Work and Codex surfaces. This plan-qualified statement does not establish the default or controls for personal Free, Go, Plus, or Pro accounts.
  3. 3. Evidence checked 2026-08-28: Claude Code lets Free, Pro, and Max users choose whether data may improve future Claude models; Team, Enterprise, API, third-party platform, and government use is not used to train generative models under commercial terms unless the customer affirmatively provides data. Feedback and optional transcript sharing follow separate rules.
  4. 4. Evidence checked 2026-08-28: Gemini Apps' Keep Activity setting controls whether future personal-account chats train Google's AI models. Turning it off stops training unless the user submits feedback, while service delivery and safety processing—including possible human review—continue.
  5. 5. Evidence checked 2026-08-28: Cursor Privacy Mode prevents Customer Data from being used for Cursor training and uses zero-data-retention agreements with model providers. Safety classifiers and triggered abuse investigations are documented exceptions, and turning Privacy Mode off permits broader storage and training use.