From Prototype to Classroom: An Intelligent Tutoring System for Quantum Education

Collection highlights systemic AI safety and governance risks across research on model adaptation, internal use, deployment, and misinformation. Key findings: (1) Fine-tuning commonly produces large, heterogeneous, and sometimes contradictory changes in measured safety — safety properties of base models do not reliably transfer to downstream adaptations, so models must be re-evaluated in deployment-relevant contexts. (2) Frontier developers’ internal use of more capable models creates distinct risks (autonomous misbehavior and insider threats); the paper proposes a harmonized internal-use risk