Jobs at Moonshot

Conduct red teaming and adversarial evaluations of AI systems across violence, extremism, child safety, grooming, crisis, and other harm categories. Design evaluation frameworks, test scenarios, scoring criteria, taxonomies, and interventions that translate real-world harm expertise into structured technical outputs. Review model responses, exercise expert judgment, and produce methodology and findings for technical and government audiences while meeting ethical, legal, contractual, and data protection obligations.