Commercial large language models are typically deployed as black-box API services, requiring users to trust providers to execute inference correctly and report token usage honestly. We present IMMACULATE, a practical auditing framework that detects economically motivated deviations-such as model substitution, quantization abuse, and token overbilling-without trusted hardware or access to model internals. IMMACULATE selectively audits a small fraction of requests using verifiable computation, achieving strong detection guarantees while amortizing cryptographic overhead. Experiments on dense and MoE models show that IMMACULATE reliably distinguishes benign and malicious executions with under 1% throughput overhead. Our code is published at https://github.com/guo-yanpei/Immaculate.
@article{arxiv.2602.22700,
title = {IMMACULATE: A Practical LLM Auditing Framework via Verifiable Computation},
author = {Yanpei Guo and Wenjie Qu and Linyu Wu and Shengfang Zhai and Lionel Z. Wang and Ming Xu and Yue Liu and Binhang Yuan and Dawn Song and Jiaheng Zhang},
journal= {arXiv preprint arXiv:2602.22700},
year = {2026}
}