Vasisht Duddu, Adam Caulfield, Xi He · 2026 · arXiv
DOI: 10.48550/arXiv.2610.27543 · open-access PDF
규정은 기업이 규제 샌드박스(regulatory sandbox)에서 AI 모델을 실행하도록 요구하지만, 이러한 샌드박스(sandbox)를 설계하기 위한 기술적 메커니즘을 정의하지는 않습니다. 이상적으로 규제 샌드박스(regulatory sandbox)는 (i) 준수(규제에 대한 검사 및 의도된 행동, 즉 허가받지 않은 도구 및 시스템 호출의 편차)를 보여야 합니다. (ii) 기밀성(regulator로부터 모델 무게를 보호한다); (iii) 인증 가능성(monitoring를 위해 모델 행동에 대한 방해 증거를 생산한다) 및 (iv) 확장 가능성(evidencability)을 보여야 합니다.
Machine translation (Meta M2M100); the English original follows.
Regulations require companies to run AI models in a regulatory sandbox, an isolated environment that enables monitoring model behavior by regulators, prior to deployment. But, they do not spell out the technical mechanisms to design such a sandbox. Ideally, a regulatory sandbox should show (i) compliance (checks against regulations, and deviations from intended behavior, i.e., unauthorized tool and system calls); (ii) confidentiality (protects model weights from regulators); (iii) attestability (produces tamper-resistant evidence of model behavior for monitoring); and (iv) scalability (evidenc
Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.27543, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange