Skip to main content
Heirloom Praxis
Heirloom Tools
No AI — fixed rules · ~4 min

Heirloom Shield Scanner

One decision, answered by rules you can read. Describe the situation — never paste the material — and it returns green, amber or red, with the reason, what to remove, what stays with a person, and a safer way to get the same thing. Its companion, the Heirloom Shield self-assessment, reviews your whole practice; this answers a single question at the moment you are asking it.

What it does

  • Classify one specific use — the material, the task, who it belongs to, whether you have authority, and where it would go
  • Escalate on the most sensitive thing involved, so a harmless-sounding task never downgrades sensitive material
  • Ask a question rather than guess when it does not have enough to rate responsibly
  • Name what would have to change for a red answer to become workable

What it does not do

  • Give legal, medical, cybersecurity or regulatory advice, or any form of certification
  • Accept your actual documents. It takes categories, never files or pasted records
  • Let a language model decide the answer. The rules are deterministic and inspectable
  • Treat an enterprise account as automatically safe, or an unknown as a green light

Step 1 of 4

0%

What you want to do

Describe the situation. Do not paste the material itself.

What are you trying to do?This tool needs no private material to answer. Describe the situation, never paste the record.
Check an AI tool

What we know about a service, and when we last looked

The scan above is about your situation. This is about the service you were thinking of using — and right now it is a shelf with the labels printed and nothing on it.

Read the tiers, not the badge

Every provider we managed to read answers the training question differently depending on which account you are on. Claude does not train on commercial inputs by default; a consumer chat can be opted in. Gemini improves its models from a personal account and does not from a school one. That is the finding, and it is why these cards refuse to give you one answer per product.

4 of 6 entries have been partly reviewed. The rest say unknown and mean it — a confident sentence about a policy nobody verified is worse than an admitted blank.

Who did this reading

The entries marked as partly reviewed were read from the providers' own published pages by an automated agent on 20 August 2026, not by a person. Every claim below links to the page it came from. Treat them as a starting point for your own reading rather than as a verified account, and check the source before anything sensitive rides on it.

  • ChatGPT

    Not reviewed

    OpenAI

    Trains on what you put in
    Unknown
    Opt-out available
    Unknown
    Stronger business controls
    Unknown
    Retention documented
    Unknown

    Not yet checked. Treat everything above as genuinely unknown rather than as a mild yes.

    openai.com refused automated retrieval on 20 August 2026, so nothing here has been read. The pages are public and a person can open them directly.

  • Claude

    Depends on your account type

    Anthropic

    Trains on what you put in
    Unknown
    Opt-out available
    Yes
    Stronger business controls
    Yes
    Retention documented
    Yes

    Commercial (Claude for Work, Team, Enterprise, API)

    Not used for training by default. Submitting feedback through the thumbs up/down button is an opt-in that allows the related conversation to be used, and an organisation owner can disable that button for their members.

    Sources (2)
    • “By default, we will not use your inputs or outputs from our commercial products… to train our models.” provider documentation, read 2026-08-20
    • Feedback conversations are stored “in our secured back-end for up to 5 years”. provider documentation, read 2026-08-20

    Consumer (Free, Pro, Max)

    Opt-in rather than automatic: chats are used to improve Claude only if model improvement is enabled. Conversations flagged for safety review may be used regardless. Incognito chats are excluded even with model improvement on.

    Sources (1)
    • Training applies where “you choose to allow us to use your chats and coding sessions to improve Claude”; “your Incognito chats are not used to improve Claude, even if you have enabled Model Improvement.” provider documentation, read 2026-08-20

    Read from Anthropic's own privacy documentation. The product-level training field is left unknown because the answer genuinely differs between the commercial and consumer tiers.

  • Gemini

    Depends on your account type

    Google

    Trains on what you put in
    Unknown
    Opt-out available
    Yes
    Stronger business controls
    Yes
    Retention documented
    Yes

    Personal Google Account

    Conversations are used to maintain and improve Google's services, including the generative AI models, and human reviewers read some of them. Turning Keep Activity off stops future chats being used for improvement, but does not stop all human review. Reviewed chats survive deleting your activity and are kept for up to three years.

    Sources (3)
    • “Human reviewers (including trained reviewers from our service providers) review some of the data we collect for these purposes.” privacy policy, read 2026-08-20
    • “Chats reviewed by human reviewers… are not deleted when you delete your activity. Instead, they are retained for up to three years.” privacy policy, read 2026-08-20
    • “Even if your Keep Activity setting is off or you use temporary chats, Google still uses your chats to respond to you and help protect Google, our users, and the public, including with help from human reviewers.” privacy policy, read 2026-08-20

    School or work Google Account

    Chats and uploaded files are not read by human reviewers and are not used to improve the models.

    Sources (1)
    • “When you use Gemini Apps with a school Google Account, your chats and uploaded files in Gemini Apps won’t be reviewed by human reviewers or otherwise used to improve generative AI models.” privacy policy, read 2026-08-20

    Read from the Gemini Apps Privacy Hub. The opt-out is real but partial, which a boolean cannot express — see the personal-account finding.

  • Microsoft Copilot

    Depends on your account type

    Microsoft

    Trains on what you put in
    Unknown
    Opt-out available
    Yes
    Stronger business controls
    Yes
    Retention documented
    Yes

    Microsoft 365 Copilot (commercial licence)

    Prompts, responses and data reached through Microsoft Graph are not used to train the foundation models. Interactions are stored as Copilot activity history under the organisation's existing Microsoft 365 commitments, and administrators can set retention policies through Purview. Customer feedback is optional and separately controlled.

    Sources (2)
    • “Prompts, responses, and data accessed through Microsoft Graph aren’t used to train foundation LLMs, including those used by Microsoft Copilot.” provider documentation, read 2026-08-20
    • “Admins can also use Microsoft Purview to set retention policies for the data related to chat interactions with Copilot.” provider documentation, read 2026-08-20

    Free consumer Copilot

    Not covered by the commercial documentation above, and not separately reviewed. Do not read the commercial commitment as applying to a free personal account.

    The commitment read here is specifically Microsoft 365 Copilot under a commercial licence. The free consumer product is a different thing and was not reviewed.

  • Perplexity

    Not reviewed

    Perplexity AI

    Trains on what you put in
    Unknown
    Opt-out available
    Unknown
    Stronger business controls
    Unknown
    Retention documented
    Unknown

    Not yet checked. Treat everything above as genuinely unknown rather than as a mild yes.

    perplexity.ai refused automated retrieval on 20 August 2026. Search results describe an AI data usage setting and an enterprise exemption, but no page was opened and read, so nothing is recorded here.

  • Gemini Notebook

    Depends on your account type

    Google — formerly NotebookLM

    Trains on what you put in
    Unknown
    Opt-out available
    Unknown
    Stronger business controls
    Yes
    Retention documented
    Unknown

    Personal Google Account

    Not used for training unless you submit feedback, in which case Google may review the full interaction. Data may still be processed to prevent fraud, abuse and technical problems. No retention period was stated on the page read.

    Sources (1)
    • “Your data is protected, and is not used to train Gemini Notebook unless you provide feedback.” provider documentation, read 2026-08-20

    Google Workspace or Workspace for Education

    Uploads, queries and responses are not read by human reviewers and are not used to train models.

    Sources (1)
    • “As a Google Workspace or Google Workspace for Education user, your uploads, queries and the model’s responses in Gemini Notebook will not be reviewed by human reviewers, and will not be used to train AI models.” provider documentation, read 2026-08-20

    Renamed by Google on 16 July 2026. Still the same standalone product at notebooklm.google, so an older link or an older policy reading is not necessarily out of date — but check which name a page is using before assuming it describes what you are using.

Whatever is recorded here

Provider terms change, and they differ by account tier and by country. Whatever is recorded here, verify the current terms yourself before putting anything sensitive through a service.