Source · Tier B · PreprintSource · Verifiable constraints on frontier training via proofs of compartmentalization
Verifiable constraints on frontier training via proofs of compartmentalization
D. Reuter, L. Marks, A. Carlucci, J. Ng, J. Petrie, J. Hausenloy, A. Karvonen, M. Baker. 2026. ICML 2026 Workshop on Technical AI Governance Research.
| Original | https://openreview.net/forum?id=ajrtHuFw2S |
|---|---|
| Archived copy | https://icml.cc/virtual/2026/78588 |
| Version | A poster paper in the Second Workshop on Technical AI Governance Research at ICML 2026, not a main-conference paper. The full 20-page PDF was assessed on 2026-10-27. Sections 4–5 specify Pod-PoComp and Task-PoComp, conditional soundness bounds and illustrative costs. Appendix C specifies the erasure check. Appendix D reports a prototype of one component, deterministic inference. No end-to-end PoComp implementation is reported. |
| Accessed | 2026-10-27 |
| Imported from | hodgkins-ai-verification-papers@c71e59ff0e8e |
| Note | Listed under "Zero-knowledge proofs" in the Hodgkins bibliography (CC BY 4.0). |