Sharing audit

Use this page to check whether your AI tools train on your work. Each section gives the provider's defaults and retention terms, with steps for changing the available settings.

The information comes from the providers' own pages, checked on 28 September 2026. Terms can change. Follow the source links if something looks different.

Claude

Anthropic: Claude apps, Claude Code, Claude API
Your planUsed for training?How long data is kept
Free, Pro, MaxYour choice
Includes Claude Code when you sign in with one of these plans.
Up to 5 years if you allow training. 30 days if you do not.
Team, Enterprise, EducationOff by defaultKept while it is in your account. Deleted chats are removed within 30 days.
API, including Claude Code with an API keyOff by defaultDeleted within 30 days.

In the Claude app or on claude.ai

  1. Select your name, then Settings.
  2. Select Privacy, or open Claude's privacy settings directly.
  3. Find Help improve our AI models and switch it on or off.

In Claude Code

  1. Type /privacy-settings and press Enter.
  2. It shows your current choice and lets you change it.

This command works on Pro and Max plans only. It is the same account setting as the one on claude.ai, so changing it in either place changes both.

What still applies with training off

  • Feedback. If you press thumbs up or thumbs down, or send a bug report from Claude Code, that whole conversation is kept for up to 5 years and may be used for training.
  • Safety review. Conversations flagged for safety review may be used to improve safety systems, and are kept for up to 2 years.
  • Training that has already happened. Turning the setting off stops future use. Work already included in a trained model stays in it.
  • Through Amazon Bedrock or Google Cloud. Those providers' own terms apply, and Anthropic's settings do not.

Sources: Anthropic's guide to the setting, consumer retention, consumer training, commercial training, commercial retention, and Claude Code data usage.

ChatGPT and Codex

OpenAI: ChatGPT, Codex, OpenAI API
Your planUsed for training?How long data is kept
Free, Go, Plus, ProOn by default
Includes Codex when you sign in with one of these plans.
Until you delete it. Deleted chats are removed within 30 days.
Business, Enterprise, Edu, Academic ResearchersOff by defaultSet by your workspace administrator.
API, including Codex with an API keyOff by default
Unless the owner of your organisation has opted in to share data.
Up to 30 days.

In ChatGPT

  1. Open your account menu and select Settings.
  2. Select Data controls.
  3. Select Improve the model for everyone, turn it off, and select Done.

The setting applies to your account on every device. You can also choose Do not train on my content in OpenAI's Privacy Portal. Either one is enough.

In Codex

  1. Turn off Improve the model for everyone in ChatGPT, as above. It covers your Codex tasks too.
  2. Then open Codex settings and check Include environments. It is a separate setting, and the ChatGPT one does not change it.

Include environments controls whether additional context from your Codex environments can be used for training. OpenAI's help page links to it in Codex's general settings.

What still applies with training off

  • Feedback. If you press thumbs up or thumbs down, the whole conversation may be used for training even though you opted out.
  • Earlier conversations. Opting out covers new conversations only.
  • Your saved chats. Turning the setting off does not delete anything. Chats stay until you delete them.
  • Temporary chats. These are not used for training, and OpenAI may keep a copy for up to 30 days for safety.
  • Your organisation's API settings. If you use an API key, the owner of the organisation may have opted in to share inputs and outputs with OpenAI. Only they can see or change that.

Sources: OpenAI's data controls guide, how your data is used, chat retention, using Codex with ChatGPT, enterprise privacy, API data guide, and API data sharing.

Gemini

Google: Gemini app, Workspace, Gemini API, AI Studio
Your planUsed for training?How long data is kept
Gemini app, personal accountOn by default
People at Google may read a sample of chats.
18 months by default. You can choose 3, 18, or 36 months.
Gemini in Google WorkspaceOff by default
Depends on your edition. Your administrator controls it.
Set by your administrator.
Gemini API and AI Studio, free tierOn
There is no switch. People at Google may read it.
Not stated by Google.
Gemini API and AI Studio, paid tierOffLogged for 55 days to detect abuse.

In the Gemini app, on the web

  1. Open Settings & help, then Activity, or open your Gemini activity directly.
  2. Find Keep Activity. If it says On, select it.
  3. Choose Turn off, or Turn off and delete activity to remove past chats as well.

For a single conversation, a temporary chat is not used for training.

For the Gemini API and AI Studio

  1. Link a billing account to at least one API project.
  2. Once billing is on, your API and AI Studio prompts fall under the paid terms and are not used for training.

In the European Economic Area, Switzerland, and the United Kingdom, the paid terms apply to everyone, including the free tier.

What still applies with Keep Activity off

  • 72 hours. Conversations are still saved with your account for up to 72 hours.
  • Chats already reviewed. Chats that a person at Google has reviewed are kept for up to three years, and deleting your activity does not remove them.
  • Feedback. If you send feedback, Google collects it along with context from the conversation.
  • Safety. Google still uses chats to protect its services and users, including with human reviewers.

Sources: Google's Gemini Apps Privacy Hub, managing Gemini activity, Workspace Privacy Hub, Gemini API terms, API billing, and API usage policies.

GitHub Copilot

GitHub and Microsoft
Your planUsed for training?How long data is kept
Free, Pro, Pro+On by default
Since 24 April 2026. Covers your prompts, code snippets, and the responses.
Not stated by GitHub.
Business, EnterpriseOffIn your editor, prompts are not kept. In chat on github.com, mobile, and the command line, 28 days.

On github.com

  1. Select your profile picture, then Copilot settings, or open your Copilot settings directly.
  2. Find Allow GitHub to use my data for AI model training.
  3. Set it to Disabled.

If you cannot find the setting

  • Your account may have a Business or Enterprise licence, where training is already off.
  • Members of a paid organisation are excluded from training.
  • Students with Copilot Student, and teachers with free Copilot Pro, are not affected.

What to know

  • Who sees it. Staff at GitHub and Microsoft who work on model development may access data collected for training.
  • What is left out. Private repositories, issues, and discussions stored on GitHub are not used. Only what passes through Copilot is.

Sources: GitHub's guide to Copilot policies, policy announcement, changelog, Copilot FAQ, and community FAQ.

Cursor

Anysphere
Your planUsed for training?How long data is kept
Hobby, Pro, Pro+, UltraCheck your setting
Allowed whenever Privacy Mode is off. Cursor does not state the default.
Not stated by Cursor.
TeamsOff
Cursor says this is enforced for every team member.
Not stated by Cursor.
EnterpriseOff by default
Privacy Mode is on by default.
Not stated by Cursor.

In Cursor

  1. Open Cursor Settings with Cmd + Shift + J on a Mac or Ctrl + Shift + J on Windows or Linux.
  2. Select General.
  3. Switch Privacy Mode on.

What still applies with Privacy Mode on

  • Your own API keys. If you connect your own key, that provider's terms apply and Cursor's protections do not.
  • Abuse detection. Prompts that trigger abuse detectors may be stored for investigation.
  • Feedback. Anything you report to Cursor as feedback may be used.
  • Cloud Agents. This feature requires Cursor to store your code.

Sources: Cursor's data use page, privacy help, enterprise data governance, security page, and privacy policy.

What these settings cover

These steps show what you have asked each provider to do under its current terms. The source pages describe the provider's commitments and the exceptions that apply.

The letter asks for a collective agreement on training permissions for research conversations and agent sessions, with fair compensation and a continuing voice for scientists.

Add your name →
Read the position paperResearch Router Co-Op →