Security analysis reveals vulnerabilities and defense conditions in large language model inference metering, highlighting risks of client-side under-payment.
This research artifact accompanies the manuscript “Token-Accounting Integrity in LLM Metering: A Systematic Study of Client-Side Under-Payment.” It contains the reproducible testbed, attack and defense implementations, experimental scripts, raw and processed results, generated tables and figures, TLA+ formal specification and model-checking outputs, and documentation required to reproduce the reported experiments. The artifact corresponds to release v1.0.1 of the accompanying GitHub repository. The study evaluates accounting integrity in LLM inference metering under an honest-provider/dishonest-client threat model, including state synchronization, commitment timing, usage authority, asynchronous settlement, and defense sufficiency conditions.
No takes yet. Share an insight, caveat, or question.
hardik . (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: