Safety & rights
As AI systems move into public benefits, employment, health and other consequential services, governments need more than a provider’s assurance that a model is safe. Negotiated obligations can establish independent evaluations, reporting of serious incidents and material model changes, limits on how sensitive data is used, and meaningful avenues for human review and appeal. The specific requirements should scale with the risks of the service, and regulators should have enough access to verify compliance. These protections help public authorities remain answerable for decisions made using imported models.
See the 4 specific obligations
- OB-01 Independent AI evaluations
- OB-02 Incident reporting & transparency
- OB-03 Privacy and data-use safeguards
- OB-04 Citizen redress and contestability
