Responsible AI
Building an adult AI product responsibly is not a marketing line. It is a set of design decisions, model behaviors, and escalation paths. This page lays out the principles we work from and how they show up in the product.
Principles
Adults only.
Strict age-gating at the door and at every step. Character design, prompts, and model behavior assume an adult audience. If you are not an adult, you are not the user we built this for.
Fiction over imitation.
Every character on the platform is a fictional adult. No real people. No deepfakes. No celebrity look-alikes. Fantasy is the product; impersonation is not.
Consent as design.
Safewords are honored. Scene negotiation and soft refusals are baked into the model rather than bolted on. The system slows down when consent cues are missing.
Zero tolerance for CSAM.
Multi-layer detection from prompt to output, an abuse queue with the highest priority, evidence preservation, and reporting to NCMEC and law enforcement where required.
Privacy as default.
Chats are private to your account. Retention is minimal. We do not sell personal information and we do not share it for cross-context advertising.
Transparency on limits.
We publish where the lines are: the Content Policy, this page, the DMCA and 2257 statements. Refusals are clear instead of evasive.
How the system is built
Safety is layered. No single check carries the whole load; each layer catches what the previous one missed.
- Character prompts: every persona is built on system prompts that fix the character as an adult and bind to platform rules. These are not user-editable.
- Input classifiers: prompts that target hard-blocked categories - minor sexualization, real-person sexual content, bestiality, weapons or hard-drug synthesis - are rejected at the door.
- Output filters: generations are reviewed before delivery; severe-category outputs are blocked, logged, and surfaced for human review.
- Human review: the trust and safety team handles reports, borderline cases, enforcement, and appeals. People decide the hard calls.
- Red-team evaluation: the model is tested continuously against adversarial prompts that try to bypass the rules.
Character design
All characters are designed and depicted as adults. Visual generation pipelines reject minor-coded outputs. Persona templates require adult markers in both art and personality. Custom personas you build follow the same rules - the system will not let you describe a character into being a minor.
User controls
Control sits with you, the adult who chose to use the product:
- End or restart a scene at any time.
- Set hard limits and safewords; the character honors them.
- Block or mute a persona.
- Report a conversation through the in-product flow or the reporting page.
- Delete chats and request account closure at any time.
Data and training
By default, your private chats are not used to train general AI models. If we ever introduce a training opt-in, it will be explicit, granular, and revocable from your account at any time. Aggregated and de-identified abuse signals are used to improve safety classifiers without retaining identifying conversation text beyond what is required to act on a report. See the Privacy Policy for the full picture.
Research and evolving safety
The model, the rules, and the legal landscape will all keep moving. We update our classifiers, our character templates, and this page as we learn. The "Last updated" date at the top reflects the current revision.
Talk to us
If you have feedback, a concern, or a safety report, email support@myaislut.net or open the reporting page. The full set of rules lives in our Content Policy.