Conditions et politiques
Acceptable use policy
What Mandi's models may be used for, and the uses we prohibit.
Last updated: 17 September 2026
This English version is provided for convenience. In case of any discrepancy, the French version prevails.
This document is a working draft. It describes what Mandi does today, but it must be reviewed by a lawyer before it binds the organisation. Passages still open are marked “to be completed”.
Why this policy
A useful model can be misused. This policy states clearly what we refuse, even before our models become widely available. It applies to any use of Mandi's models, whatever the interface: website, platform, programming interface or distributed weights.
Prohibited uses
Harm to people. Producing content that harasses, threatens, defames or exposes a person; generating sexual content involving minors; producing images or voices imitating a real person in order to deceive.
Information manipulation. Fabricating false information presented as verified; producing automated political propaganda; impersonating a public institution, a media outlet or a company.
Fraud and crime. Preparing a scam, identity theft or phishing; producing forged administrative documents; facilitating money laundering or human trafficking.
Security threats. Designing malicious software; preparing a computer intrusion; helping to design weapons, including chemical, biological or explosive weapons.
Automated decisions about people. Refusing employment, credit, housing, care or a social benefit on the sole basis of a model's output, without human review.
Circumventing safeguards. Bypassing a model's safeguards, or helping others to do so.
Supervised uses
Some fields call for particular care and human review: health, law, education, public administration, employment and finance. In these fields, a Mandi model may assist a competent person, never replace them.
What we expect from you
- Tell your users that they are interacting with a model, not with a person.
- Check the information produced before sharing it.
- Report any failures you observe to contact@mandia.org.
What we do on our side
- Publish the capabilities, limitations and safety evaluations of each model before it is opened.
- Document the origin of the training data and the consents obtained.
- Restrict or suspend any access used in breach of this policy.
Reporting
Misuse, problematic content or a security flaw can be reported to contact@mandia.org. We acknowledge receipt of and handle every report.
A question about this document? contact@mandia.org