Track trials, findings, and clinical evidence as it lands.
Search papers, labs, and topics across Lattice.
FinHarness is introduced as an inline safety harness for finance LLM agents, designed to prevent unauthorized actions and ensure legitimate workflow execution. It employs a Query Monitor, Tool Monitor, and Cascade module to assess risk at each step and adaptively route verification between lightweight and advanced LLM judges. By re-injecting identified risk factors into the agent's input, FinHarness enables the agent to self-correct, resulting in a significant reduction in Attack Success Rate (ASR) from 38.3% to 15.0% while maintaining benign approval rates and reducing the computational cost of advanced LLM judge calls by 4.7x.
Finance LLM agents can now block unauthorized actions mid-trajectory without sacrificing performance, thanks to a novel inline safety harness that adaptively routes verification between lightweight and advanced LLM judges.
Finance LLM agents must simultaneously block prompt-induced unauthorized actions and approve legitimate multi-step business workflows. However, boundary filters often miss irreversible mid-trajectory tool calls, while post-hoc LLM judges perform auditing only after termination -- too late for intervention and at a computational cost that scales linearly with trace length. We present FinHarness, an inline safety harness that wraps a finance agent end-to-end with three components: a Query Monitor that fuses single-turn intent with cross-turn drift, a Tool Monitor that evaluates each prospective tool call, and a Cascade module that integrates per-step risk and adaptively routes verification between a lightweight and an advanced-tier LLM judge. Fired risk factors are re-injected into the agent input as ex-ante evidence, enabling the agent to refuse, re-plan, or approve on its own. On FinVault, routed FinHarness cuts ASR from 38.3% to 15.0% while largely preserving benign approval ($41.1\% \to 39.3\%$), and uses $4.7\times$ fewer advanced-judge calls than an always-advanced ablation.
Tsinghua AI