Rapid AI development across industries raises pressing security and privacy risks. This work presents a unified comparison of large language models, AI agents, and embodied agents, introducing a taxonomy of risks spanning data, models, systems, content, and applications, alongside a catalog of 24 specific threats. We contrast attack surfaces and methods across the three system types to reveal common patterns and distinctive vulnerabilities. We also survey mainstream AI security assessment frameworks and evaluate how relevant laws and regulations currently address these risks. Finally, we outline concrete directions for future research and practice aimed at building robust, secure agent ecosystems.
Modern large language models (LLMs) rely on reinforcement learning to build strong capabilities in individual domains, but integrating those capabilities into a single deployable model remains challenging. By routing each sample to the teacher whose domain matches it, existing approaches let a domain label decide which teacher provides supervision. However, domain expertise holds only on average: the matched teacher is not always correct on a given sample, while a teacher from another domain sometimes is. The reliable teacher therefore has to be identified per sample, not per domain. In this paper, we introduce Multi-Teacher Self-Distillation Policy Optimization (MT-SDPO), an on-policy distillation method that unifies several frozen teachers into one student model. MT-SDPO consists of three components: (1) self-anchors, where a rollout is supervised by a correct rollout from its own group; (2) answer-verified eligibility, where a teacher may supervise a sample only if its own answer passes a verifier; and (3) privileged distillation, which merges the anchor and all verified feedback into one context that an exponential moving average self-teacher reads and the student does not, thereby keeping one policy at deployment. Across five students from three model families, MT-SDPO lifts the weakest domain of Qwen3-8B by 14.79 points and narrows its domain gap by 74.7%, a better balance than serving one matched teacher per domain. Verified reliability, not domain membership, should decide who teaches. Code is available at https://github.com/hexixiang/MT-SDPO.
Xi-Xiang He, Xing-Ming Li, Baiqi Wu et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
