Red Flags

Red Flags answers one practical question: how to recognize unsafe privacy pressure impersonation and manipulative AI claims. Start with a bounded result rather than a promise that any app will understand you perfectly. A clear result can be reviewed; a vague wish for flawless companionship cannot.

Keep privacy, consent, account controls and personal wellbeing ahead of momentum or immersion. This matters because generated dialogue can sound confident while still being improvised. Treat a strong response as part of the scene, not proof of memory, expertise, consent or feeling.

The public evidence used for this page was checked on October 2, 2026. It covers U.S. Federal Trade Commission: privacy and security evidence framework; OWASP Foundation: generative-AI privacy, disclosure and manipulation risks. These materials describe current claims, rules or risk frameworks; they do not prove what a particular account stores or how a future model update will behave.

Before you begin, make a compact scene card for “Red Flags.” Include the fictional character name, adult age, setting, relationship premise, allowed themes, excluded themes, stop phrase and desired ending. Do not copy a real person’s identity, private messages or likeness into the card.

Use specific observable instructions. Instead of asking for “better chemistry,” name the pace, length, tone and continuity detail you want. Instead of asking the character to “always remember,” list the two or three facts that matter in this scene and expect to restate them when the product does not retain them.

Keep consent reversible inside the fiction. A boundary written at setup is not a one-time permission slip. Pause when the scene changes direction, when a generated reply introduces an excluded theme, or when you no longer want to continue. Editing or leaving a scene is a normal creative choice.

Protect the real person behind the story. Use a separate fictional biography, avoid unique personal details, and check whether exports, deletion requests and subscription controls are explained before sharing sensitive material. A private-feeling interface is not the same as a private conversation.

Evaluate the result with a short record: what the app was asked to do, what it actually did, which control was used, and whether the outcome can be reproduced. This is more useful than declaring a platform “best” after one unusually good or bad exchange.

If the scene creates pressure, distress or confusion about what is real, stop the product workflow. Talk to a trusted person or qualified professional when human support is needed. An AI character is entertainment software, not a therapist, emergency service or person who can consent.

Finish by choosing the next unresolved step, not by opening every comparison at once. For Red Flags, completion means you have a usable decision or scene rule, know what remains uncertain, and can point to the current public source that supports each product or safety claim.