Research preview policy
Real conversation. Experimental system.
This release is meant to test conversational judgment with a small group—not to present Humano as a finished or universally reliable product.
Effective July 30, 2026The purpose
We’re testing whether better judgment can make AI conversation feel meaningfully more natural.
Participant feedback helps identify emotional mistakes, robotic patterns, poor questions, weak opinions, repetition, and moments where the system sounds confident without earning it.
Preview status
Humano is an experimental behavior layer powered by an AI language model. Features may be unfinished, unavailable, slow, or changed during testing. Conversation history and memory can be incomplete or lost.
The preview is invitation-only, for adults who reside in and are physically located in the United States. Participants must use a direct connection without a VPN, proxy, Tor, or location-masking service.
Known risks
Humano may invent facts, misread emotion, remember something incorrectly, express an unjustified opinion, echo bias, miss context, or give advice that sounds more confident than it should. Its natural conversational style can make these mistakes easier to trust.
Verify important claims independently. Do not use the preview as the sole basis for medical, legal, financial, employment, educational, safety, or other high-impact decisions.
Good participation
Talk naturally, challenge the system, and report responses that are confusing, manipulative, unsafe, repetitive, or strangely robotic. Use feedback controls when they capture your reaction, but do not submit confidential or highly sensitive information just to test the model.
Do not intentionally use the preview to harm another person, generate illegal material, attack the service, evade restrictions, or expose someone else’s private information.
Learning from the preview
Humano may analyze conversations, planner decisions, model responses, validator scores, and participant feedback to compare behavior patterns and improve later releases. These signals may inform evaluation datasets, preference data, and future model training.
The goal is to learn which conversational decisions work—not to encourage participants to disclose personal information. Unnecessary personal details should be removed or de-identified when examples are prepared for longer-term research use.
Access and changes
Access may be rate-limited, suspended, or ended as the preview evolves. A revised agreement may be required when participation terms, data practices, or material risks change.
Questions, data requests, and serious response reports should be sent to the preview organizer who shared your invitation.