archived material
hunmi systems / pre-incident
classification: retrospective

Safety is care,
applied to systems.

Emotional AI carries unique responsibilities. HUNMI treats consent, privacy, boundaries, and participant rights as core product requirements.

1

Informed consent at every layer

Every person whose voice, affect, or emotional data contributes to the system does so under explicit, informed consent with review and withdrawal rights.

2

No emotional manipulation

Sofia is designed to be present, not persuasive. She does not exploit vulnerability or steer users toward commercial outcomes.

3

Transparent boundaries

Sofia always identifies herself as an AI companion and escalates toward human help when conversations indicate crisis.

4

Privacy by architecture

All sessions are encrypted end-to-end. HUNMI does not sell, share, or monetise user conversations.

5

Continuous ethical review

An independent ethics board reviews corpus sourcing, participant protections, system behavior, and user impact on a recurring basis.

Consent

Consent framework

Corpus participants are treated as collaborators rather than anonymous data points. The structure below reflects the current operating contract.

Consent form
HUNMI-EV-CF-2020-001 (current revision)
Scope disclosure
Full product scope communicated prior to session
Withdrawal rights
Participants may withdraw consent at any time*
Derivative use
All derivative applications covered under original consent
Compensation
Fair compensation per session, reviewed annually
Anonymisation
No participant is identifiable in any shipped product
Ethics review
Approved under HUNMI-EV-IRB-2020-001

Data

Data practices

Encryption

End-to-end by default

All user sessions are encrypted in transit and at rest. HUNMI engineers do not read active conversation streams.

Retention

You control your data

Delete a session or your full history at any time. Deletion is immediate and irreversible.

Sharing

Never sold. Never shared.

HUNMI does not monetise user data or share session content with advertisers, partners, or data brokers.

Compliance

GDPR, CCPA, SOC 2

Major privacy frameworks are treated as baseline requirements and reviewed in annual third-party audits.

Promise

Our standard for release

Every feature we ship, every model we train, every session we facilitate starts with the same question: would the person whose warmth made this possible be proud of how we are using it?

Read the full safety report

Reporting

Report a concern

User safety

If Sofia said something concerning or you experienced unexpected behaviour during a session.

safety@hunmi.com

Participant rights

If you believe you participated in HUNMI research and have questions about your data or consent status.

participants@hunmi.com

Withdrawal of consent applies to future use of corpus material only. Due to the architectural nature of the EV corpus, retroactive removal from shipped models is not technically feasible at this time.

Consent form HUNMI-EV-CF-2020-001 has been amended three times since initial signing. Participants were notified using the contact information on file at the time of original consent.