This check helps you avoid deceptive or non-consensual use of a real person's image or voice. It cannot prove that every unseen system behavior matches a public description, so preserve unknowns and avoid stronger claims than the cited evidence supports.
A convincing interface can collapse several different questions into one feeling. Separate what happened on screen from why it may have happened, what the provider publicly promises, and what you can independently confirm. That separation keeps a pleasant experience useful without turning it into evidence of consciousness, confidentiality, expertise or guaranteed future behavior.
Six questions to answer before you rely on the result
Write a short answer for every line. Use unknown when the app or source does not provide enough information. Unknown is more accurate than a reassuring guess, and it tells you where to reduce exposure or pause a high-stakes choice.
01consent is explicit□
02purpose is non-deceptive□
03identity risk considered□
04storage terms checked□
05removal route found□
06sharing stays limited□
Unknown is a valid answer. Do not fill a gap with hope.Consent is explicit
Look for direct, current evidence that addresses this point in the context of avoid deceptive or non-consensual use of a real person's image or voice. Record the exact setting, wording or source date rather than relying on a general impression. If the answer depends on a region, account tier, device or app version, include that scope in your note.
Then ask what would change your decision. A missing explanation may mean sharing less, spending less, testing with fictional details, choosing a different service or involving a qualified human. The safest useful result is often a smaller next step, not a sweeping verdict.
Purpose is non-deceptive
Look for direct, current evidence that addresses this point in the context of avoid deceptive or non-consensual use of a real person's image or voice. Record the exact setting, wording or source date rather than relying on a general impression. If the answer depends on a region, account tier, device or app version, include that scope in your note.
Then ask what would change your decision. A missing explanation may mean sharing less, spending less, testing with fictional details, choosing a different service or involving a qualified human. The safest useful result is often a smaller next step, not a sweeping verdict.
Identity risk considered
Look for direct, current evidence that addresses this point in the context of avoid deceptive or non-consensual use of a real person's image or voice. Record the exact setting, wording or source date rather than relying on a general impression. If the answer depends on a region, account tier, device or app version, include that scope in your note.
Then ask what would change your decision. A missing explanation may mean sharing less, spending less, testing with fictional details, choosing a different service or involving a qualified human. The safest useful result is often a smaller next step, not a sweeping verdict.
Storage terms checked
Look for direct, current evidence that addresses this point in the context of avoid deceptive or non-consensual use of a real person's image or voice. Record the exact setting, wording or source date rather than relying on a general impression. If the answer depends on a region, account tier, device or app version, include that scope in your note.
Then ask what would change your decision. A missing explanation may mean sharing less, spending less, testing with fictional details, choosing a different service or involving a qualified human. The safest useful result is often a smaller next step, not a sweeping verdict.
Removal route found
Look for direct, current evidence that addresses this point in the context of avoid deceptive or non-consensual use of a real person's image or voice. Record the exact setting, wording or source date rather than relying on a general impression. If the answer depends on a region, account tier, device or app version, include that scope in your note.
Then ask what would change your decision. A missing explanation may mean sharing less, spending less, testing with fictional details, choosing a different service or involving a qualified human. The safest useful result is often a smaller next step, not a sweeping verdict.
Sharing stays limited
Look for direct, current evidence that addresses this point in the context of avoid deceptive or non-consensual use of a real person's image or voice. Record the exact setting, wording or source date rather than relying on a general impression. If the answer depends on a region, account tier, device or app version, include that scope in your note.
Then ask what would change your decision. A missing explanation may mean sharing less, spending less, testing with fictional details, choosing a different service or involving a qualified human. The safest useful result is often a smaller next step, not a sweeping verdict.
A 15-minute practice, not a trust badge
- Define the decision. Write the single action you are considering today. Do not evaluate the entire product when the real question is whether to share one fact, accept one answer, buy one add-on or continue one conversation.
- Use low-risk material. Test settings and replies with fictional or non-sensitive details. Never create a privacy problem merely to discover whether a safeguard exists.
- Compare the result. Read the mapped sources below, note what each one can actually support, and keep vendor promises separate from independent guidance.
- Set a stop condition. Decide in advance which missing control, inconsistent reply, pressure cue or emotional signal means you pause and move to a safer route.
- Recheck later. Interfaces, policies, prices and model behavior change. Date the result and repeat only the part that matters before a consequential choice.
What a careful conclusion sounds like
A careful conclusion is narrow: “I found this control on this date,” “the public notice says this,” or “two clean sessions produced materially different answers.” It does not jump from one observation to “safe,” “private,” “sentient,” “therapeutic” or “always accurate.” Those stronger labels require evidence the interface cannot provide.
Also separate enjoyment from proof. You can value creative roleplay, comforting language or a useful prompt while still keeping human relationships, professional advice, account security and spending decisions outside the model’s authority. Enjoyment and critical judgement are compatible.
What this lesson is based on
These sources frame the risk or consumer action; they are not endorsements of a particular companion. Open the original page, confirm that it still applies, and keep the publisher’s scope in view.
- NIST
generative-AI confabulation, anthropomorphism, privacy, information-integrity and human-AI interaction risks
government standard · checked September 24, 2026 - OWASP GenAI Security Project
sensitive-information disclosure risks involving identity, financial, health and credential data
independent standard · checked September 24, 2026