Financial trading agents face serious risks from market shocks and attacks
SoK: Trading Agents or Market Crashers? Dissecting Robustness and Security Failures in Academic Financial LLM Trading Schemes
Cryptography and SecurityArtificial IntelligenceMultiagent Systems
Summary
Financial trading programs powered by large language models can directly influence real money in fast-moving markets. The authors found that many academic trading systems are not strong enough to handle sudden market crashes or targeted attacks from bad actors. Small errors in these systems can cause big problems like market-wide crashes. The study introduced a testing framework called FARSIGHT to evaluate how well these systems handle tough situations and potential security threats.
What this means in practice
- •For financial software developers: Use the FARSIGHT framework to test trading agents’ stability and security before deployment in volatile markets.
- •For cybersecurity teams in finance: Identify and mitigate realistic attack vectors against autonomous trading agents to prevent market disruptions.
Authors
Mengxiao Wang, Nitesh Saxena
Abstract
Autonomous large language model (LLM) agents are moving rapidly into high-stakes domains, yet existing agentic-AI security studies remain largely domain-agnostic and overlook the distinctive, high-consequence attack surface such settings create. We examine this gap through financial trading agents, a representative case of high-stakes agentic security, where a single compromised agent has direct execution authority over real capital in an adversarial, reflexive market. To this end, we present FARSIGHT (Financial Agent Robustness and Security Investigation and Global Holistic Testing), a framework that performs scheme-level evaluation of financial LLM agents on two axes: robustness under market turbulence (including flash-crash-like scenarios), and security against three attack types: attacks on information sources, attacks on agents, and agent-as-attacker behaviors. Applying FARSIGHT to 15 representative academic schemes, we find that most overlook robustness and realistic adversarial threats: 80% fail at least one core robustness metric and 100% exhibit security vulnerabilities. These two failure modes are inseparable: a small misjudgment can cascade into a market-wide crash on its own, while an adversary can deliberately trigger the same collapse at minimal cost.