IMMACULATE: A Practical LLM Auditing Framework via Verifiable Computation
Yanpei Guo, Wenjie Qu, Linyu Wu, Shengfang Zhai, Lionel Z. WANG, Ming Xu, Yue Liu, Binhang Yuan, Dawn Song, Jiaheng Zhang · 2026
Read IMMACULATE: A Practical LLM Auditing Framework via Verifiable Computation by Yanpei Guo, Wenjie Qu, Linyu Wu, Shengfang Zhai, Lionel Z. WANG, Ming Xu, Yue Liu, Binhang Yuan, Dawn Song, Jiaheng Zhang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-22700.txt.
Description
Yanpei Guo and nine co-authors audit black-box language-model services without trusting the provider. Commercial models run as APIs, so users must take correct execution and honest token counts on faith. That faith invites quiet cheating: model substitution, quantization abuse, and token overbilling that pad bills.
IMMACULATE checks a small slice of requests with verifiable computation, catching economically motivated deviations with no trusted hardware and no peek inside the model. Cryptographic overhead stays amortized across the audited fraction. Tests on dense and mixture-of-experts models show it cleanly separates benign from malicious runs at under one percent throughput overhead, giving users a practical lever against provider misbehaviour.
Catalog record
| CREATOR | Yanpei Guo, Wenjie Qu, Linyu Wu, Shengfang Zhai, Lionel Z. WANG, Ming Xu, Yue Liu, Binhang Yuan, Dawn Song, Jiaheng Zhang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 54 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-22700