Source · Tier C · Blog / articleSource · Verifiable Training of AI Models
Verifiable Training of AI Models
A. Aguirre, R. Millet. 2024. Future of Life Institute.
| Original | https://futureoflife.org/ai/verifiable-training-of-ai-models/ |
|---|---|
| Version | The page shows publication on 2024-07-23 and an update on 2025-05-20. |
| Accessed | 2026-09-25 |
| Authored or published by | Future of Life Institute |
| Note | FLI (Aguirre) and Mithril Security (Millet) describe AICert, a proof of concept that uses Trusted Platform Modules to bind a model's weights to its training code and data, and state its limits (fine-tuning only, no third-party audit, no detection of poisoned models or datasets). Code at https://github.com/mithril-security/aicert, whose README warns that AICert is still under development and not for production use (read 2026-09-25). Supports "FLI reports" statements only. |