📄 中文摘要
随着企业将人工智能代理嵌入交易系统、财务平台和金融工作流程，这些概率系统悄然获得了移动真实资金的能力。当前的讨论主要集中在偏见和幻觉问题上，但更大的问题在于架构层面：一旦大型语言模型（LLM）能够提交订单、调整限额或触发再平衡，风险面就变成了一个实时交易端点。同时，针对公共应用的攻击年增长44%，超过一半的漏洞无需身份验证。这种现状使得金融系统面临潜在的安全隐患。
人工智能代理、金融API与未经授权股票交易的隐形威胁
📄 English Summary
Ai Agents Financial Apis And The Invisible Threat Of Unauthorized Stock Trades
As enterprises integrate AI agents into trading stacks, treasury platforms, and financial workflows, these probabilistic systems are quietly gaining the ability to move real money. While the debate often focuses on bias and hallucinations, a more significant concern lies in the architectural aspect: once a large language model (LLM) can submit orders, adjust limits, or trigger rebalancing, the risk surface transforms into a live trading endpoint. Concurrently, attacks on public-facing applications have surged by 44% year over year, with over half of vulnerabilities requiring no authentication. This situation poses potential security risks to financial systems.