AI summaryⓘ
The authors created SolarChain-Eval, a testing setup that checks how well AI agents perform and behave when managing decentralized energy markets, including safety and trust concerns. It treats the market like a game where AI agents make hourly choices, and measures things like fairness, safety, and how smooth their actions are. They added a language-based planner and auditor to help catch risky decisions and keep records of interventions. Their experiments show that while AI agents can improve market performance, they may still act unsafely if not properly constrained, and the planner/auditor helps but can't fix all problems. The authors stress that testing trustworthy AI needs physical rules and clear logs of decisions and fixes.
agentic AIdecentralized energy marketsMarkov Decision ProcessGymnasiumreinforcement learningLLM Planner/Auditormarket utilityphysical safetyreward functionauditability
Abstract
As agentic AI systems are increasingly applied to cyber-physical environments, their evaluation requires assessment of both task performance and trustworthiness. In decentralized energy markets, autonomous agents may improve market utility, but may also exploit invalid physical data, create artificial liquidity, and produce unstable governance decisions. Therefore, we propose SolarChain-Eval, a physics-constrained benchmark for evaluating trustworthy economic agents. It formulates market governance as a Gymnasium-compatible Markov Decision Process, where agents make hourly decisions. SolarChain-Eval evaluates each policy across multiple dimensions, including market utility, physical safety, slippage, action smoothness, spatial fairness, and auditability. To support agentic evaluation, SolarChain-Eval incorporates an LLM-based Planner/Auditor layer. The Planner defines episode-level action bounds and audit rules, while the Auditor reviews and revises high-risk actions. All interventions are recorded through structured logs, including trigger signals, proposed actions, revised actions, and audit rationales. Experiments with static, random, myopic, RL, and RL+LLM policies reveal a clear utility-safety trade-off. RL agents improve market utility but can still produce unsafe behavior. When the physics penalty is removed, reward-maximizing agents exploit invalid generation and increase artificial liquidity. The LLM Planner/Auditor improves auditability and mitigates selected risks, but it cannot fully compensate for a misspecified reward function. These results indicate that trustworthy agentic AI evaluation requires both physical constraints and transparent intervention traces. We release data and code as open access on GitHub for replicability.