arXiv · 2611.16603
Regime-Uncertain Data Transactions: A Game-Theoretic Framework for Self-Adapting and Self-Diagnosing Multi-ML-Agent Systems
Abstract
Large firms use autonomous machine learning agents that communicate with each other across legal and geographic boundaries. The rules governing a particular transfer are fluid. They depend on where the data subject lives, where the agents operate, who their owners and targets are, and how many different versions of the rules are in effect. Descriptions of the data using tags can be inaccurate, and the counterpart agent may exaggerate its security measures.We model such a transaction between two ML agents as a regime-uncertain transfer game, with a third player acting as the enforcer. The set of rules is represented as hard constraints with a strictest-wins logic. Both agents maintain Bayesian beliefs about which rules apply. They can spend money to gain access to security measures and verify the counterpart's claims, and then negotiate within the remaining legal space using Legally-Constrained Nash Bargaining (LCNB).Using the Monotonic Concession Protocol with Zeuthen's rule, we design LCNB. We prove four properties of the model and implement it in a simulated setting with GDPR, DPDP, CCPA/CPRA, and a fictional stricter regime.Across three stress tests where residence and citizenship are at odds, the citizenship-based rule is violated in 21 to 37 percent of shared data records, while LCNB results in a violation rate of less than 1 percent.There are limitations to our results. With our default economics, LCNB shares virtually all aggregates, while having a low rate of violations partly due to this.
Explore related subjects
Keep this discovery
Explore connections, maps & timelines
Chethana Prasad Kabgere. 2026-11-14. Regime-Uncertain Data Transactions: A Game-Theoretic Framework for Self-Adapting and Self-Diagnosing Multi-ML-Agent Systems. https://searcharxiv.com/abs/2611.16603
Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite Searcharxiv by Indiciae: BibTeX · RIS.