How to Object to CHAI Using Chats for AI Training

1. Write down your exact scope

Expected result: one processing purpose named clearly enough for a written decision.

Use the wording 'personal data derived from my in-app conversations for model improvement and fine-tuning'. Note whether you are asking about future conversations, already-held conversation-derived data, or both. Asking about both does not establish what can be stopped retrospectively.

Ask the provider to separate model improvement from account operation and security. Do not describe this as withdrawing training consent: the policy identifies legitimate interests for this purpose.

Keep a private note of the policy URL, update date and relevant heading. It makes a later policy change easier to distinguish from the version behind your request.

2. Explain genuine personal grounds

Expected result: a short factual explanation of the impact on you without new intimate disclosures.

Describe what is true about your situation and why broader model improvement concerns you. Suggested wording, only if accurate: 'My conversations contain distinctive details about my work and personal relationships. Reuse beyond my own conversations creates a confidentiality concern for me, even without my account name.'

Adapt the explanation to your actual use. General categories and the practical impact are preferable to names, diagnoses, addresses or quotations. Do not invent hardship, claim professional secrecy that does not apply, or paste someone else's grounds.

Check that a reader can understand the impact without seeing the private event itself. Clarify without adding a transcript.

3. Email a minimal request to CHAI

Expected result: a sent request to a verified provider address, not an opt-out setting changed.

The policy's contact details list dpo@chai-research.com and hello@chai-research.com. Choose one initially. Identify your existing account with its associated email or known account identifier; do not attach identity documents unsolicited.

Suggested subject: 'Objection: conversation-derived model improvement'. Suggested message: 'Please review my objection to using personal data derived from my CHAI conversations for model improvement and fine-tuning. Account: [minimal identifier]. Scope: [future use, already-held data, or both]. Personal grounds: [truthful concise explanation]. Please confirm receipt and your written decision, scope and effective date. If you decline, explain the grounds. Please clarify whether ordinary service access can continue.'

Replace placeholders before sending. Where relevant, request restriction while the objection is assessed; the policy includes this possibility. Record that separately as requested, not granted.

4. Record receipt and the applicable clock

Expected result: an evidence log distinguishing sent, received, pending and decided.

Save your outgoing message, recipient, timestamp, any bounce notice, acknowledgment and ticket number. A sent-folder entry is not proof of receipt; an acknowledgment is not acceptance. If delivery fails, verify the address and try the other listed contact, retaining the failed attempt.

The ICO's right-to-object guidance is UK-specific, not a worldwide CHAI service promise. It describes one calendar month to respond, with up to two additional months for complex requests and an explanation. If proportionate identity checks are needed, it says the month starts once ID is received. Clarification may also be requested.

Label any follow-up date with its basis: applicable UK guidance or your requested date. CHAI's rights wording alone does not establish a universal deadline.

5. Read the decision for actual coverage

Expected result: a clear decision or a precise follow-up question, not assumptions about hidden processing.

Look for the account covered, purpose, effective date and treatment of future versus already-held data. Does the reply accept, reject with reasons, or merely acknowledge? Put unresolved questions in one follow-up within the same thread.

Suggested question: 'Does your response cover future conversation-derived model-improvement use, and what happens to data already prepared for that purpose?' Do not convert a reassuring answer into a claim that trained models have forgotten earlier chats.

UK guidance allows refusal in some circumstances. If unresolved and UK law applies, complain to the provider first; the ICO recommends that opportunity before escalation. Elsewhere, seek guidance from the relevant authority rather than assuming ICO jurisdiction.

6. Check privacy, payment and whether to stop

Expected result: a deliberate continue, pause or exit decision based on what is confirmed.

Do not upgrade to pursue the objection. Keep payment separate: check your existing renewal date and cancellation route if you plan to leave. Neither the request nor its receipt establishes cancellation, a refund or continued access.

CHAI's Safety Center describes an 18+ service, reporting, filters and real-time self-harm scanning. Those safety measures do not establish a training opt-out. This workflow also does not verify any in-app training switch.

Pause chats if uncertainty exceeds your comfort level. Stop if support seeks passwords, verification codes or sensitive transcripts; verify through the official contact. Continuing without intimate details does not prove training has stopped. An offline private journal is an alternative needing no companion subscription.

Summary: keep the request and outcome separate

A useful stopping point is a documented decision and an informed choice about further use.

Name the purpose, explain genuine grounds, send minimal identifiers and retain correspondence. This is not complete deletion or proof of retrospective unlearning. CHAI's deletion policy has retention exceptions. Choose whether to continue based on the response, not silence or an acknowledgment.

Frequently asked questions