Anthropic 已同意向主要财政部和中央银行简报其最新 AI 模型 Claude Mythos Preview 所识别的全球金融体系中的网路漏洞。此举是在英格兰银行行长、金融稳定委员会（FSB）主席 Andrew Bailey 要求 Anthropic 向 FSB 成员解释该模型能力之后作出的。FSB 是一个与 G20 相关的全球监管机构，成员包括美国、英国、加拿大、法国、德国、日本、沙乌地阿拉伯、澳洲和中国等主要经济体的官员，许多成员日益担心 Mythos 及类似的 AI 系统可能暴露银行网路防御中的弱点。

Anthropic 上月表示，Mythos 已发现数千个高严重性漏洞，包括所有主要作业系统和网页浏览器中的问题，这引发了外界对经济体、公共安全和国家安全可能遭受严重冲击的担忧。由于担心被滥用，该模型仅向少数组织发布，且主要在美国；但约 40 个组织，包括 Amazon、Microsoft 和 JPMorgan Chase，已获得存取权，以便修复其发现的缺陷。Anthropic 也同意向一些非美国组织提供高层简报，包括欧盟委员会，同时在白宫提出要求后，拒绝更广泛地扩大分发。

FSB 正在准备一份关于在金融业采用 AI 的良好做法报告，并计划于下个月发布征询意见版本；同时，监管机构正敦促银行和其他机构加强网路安全，并因应前沿 AI 驱动的攻击而更快部署软体修补程式。英国财政部和金融监管机构已告知伦敦金融城机构采取积极措施降低风险，IMF 则警告，最新模型可能把网路风险提升为潜在的宏观金融冲击。IMF 也表示，网路风险不受国界限制，而新兴和发展中国家往往防御较弱、资源较少，可能因此承受不成比例的暴露；不过，一些当局怀疑在地缘政治紧张局势下是否会出现协调一致的全球回应。

Anthropic has agreed to brief leading finance ministries and central banks on cyber vulnerabilities in the global financial system identified by its latest AI model, Claude Mythos Preview. The move followed a request from Andrew Bailey, governor of the Bank of England and chair of the Financial Stability Board (FSB), for Anthropic to explain the model’s capabilities to FSB members. The FSB is a G20-linked global watchdog that includes officials from major economies such as the US, UK, Canada, France, Germany, Japan, Saudi Arabia, Australia and China, and many members are increasingly concerned that Mythos and similar AI systems could expose weaknesses in banks’ cyber defences.

Anthropic said last month that Mythos had found thousands of high-severity vulnerabilities, including issues in every major operating system and web browser, raising concerns about severe fallout for economies, public safety and national security. The model has been released to only a limited number of organisations, mostly in the US, because of fears it could be misused, but about 40 organisations, including Amazon, Microsoft and JPMorgan Chase, have been given access so they can fix the flaws it identifies. Anthropic has also agreed to provide high-level briefings to some non-US organisations, including the European Commission, while declining to expand distribution more broadly after a request from the White House.

The FSB is preparing a report on sound practices for adopting AI in finance and plans to release it for consultation next month, while regulators are urging banks and other institutions to strengthen cyber security and deploy software patches faster in response to frontier AI-driven attacks. The UK Treasury and financial regulators have told City of London institutions to take active steps to reduce risk, and the IMF has warned that the latest models could elevate cyber risk to a potential macro-financial shock. The IMF also said cyber risk does not respect borders and that emerging and developing countries, often with weaker defences and fewer resources, may be disproportionately exposed, although some authorities doubt a coordinated global response will emerge amid geopolitical tensions.