Benchmark reveals agentic AI models fail repeatedly at accounting tasks

AgentRelBench, a new reliability framework, tested nine LLM models on accounting operations and found systematic failures in irreversible tasks—exposing that single-run audits miss critical agent d...