- Red team and conduct adversarial evaluation of AI systems against defined harm categories.
- Review model responses against harm and risk criteria and provide expert judgement.
- Apply subject matter expertise in areas like CSEA, radicalisation, crisis, or teen safety.
- Design evaluation frameworks that translate real-world harm knowledge into structured, testable criteria.
- Develop intervention logic to connect at-risk users to support resources.
- Draft methodology or findings suitable for technical and government audiences.