#compliance
Regulatory compliance, policy enforcement, and audit trail research for financial AI systems
FinToolBench pairs 760 live financial API tools with 295 executable queries to benchmark LLM agents on real financial tasks — revealing that GPT-4o's conservative 22.7% invocation rate yields higher answer quality (CSS 0.670) than Qwen3-8B's aggressive 87.1% TIR, while intent mismatch exceeds 50% across every model tested.
CMU and NC State researchers propose using System-Theoretic Process Analysis (STPA) and a capability-enhanced Model Context Protocol to derive formal safety specifications for LLM agent tool use, with Alloy-based verification demonstrating absence of unsafe flows in a calendar scheduling case study.
FinAuditing tests 13 LLMs zero-shot on 1,102 real SEC XBRL filing instances; top scores are 13.86% on financial math verification and 12.42% on concept retrieval—results that directly bound what AI accounting tools can be trusted to automate without external tooling.
AGrail (ACL 2025) introduces a two-LLM cooperative guardrail that adapts safety checks at inference time via test-time adaptation, achieving 0% prompt injection attack success and 95.6% benign action preservation on Safe-OS — compared to GuardAgent and LLaMA-Guard blocking up to 49.2% of legitimate actions.
ShieldAgent (ICML 2025) replaces LLM-based guardrails with probabilistic rule circuits built on Markov Logic Networks, achieving 90.4% accuracy on agent attacks with 64.7% fewer API calls — and what it means for verifiable safety in financial AI systems.
AuditCopilot applies open-source LLMs (Mistral-8B, Gemma, Llama-3.1) to corporate journal entry fraud detection, cutting false positives from 942 to 12 — but ablation reveals the LLM functions primarily as a synthesis layer on top of Isolation Forest scores, not as an independent anomaly detector.
Anthropic's Constitutional AI paper (Bai et al., 2022) trains LLMs to follow rules using AI-generated feedback rather than human harm labels. This research log examines how the RLAIF critique-revise-preference pipeline maps onto write-back safety for autonomous Beancount ledger agents — and what Goodharting, calibration failures, and dual-use risks look like when the "constitution" is a chart of accounts instead of an ethics ruleset.
