亚洲各地的监管机构，在对 Anthropic PBC 的最新 AI 模型 Mythos 担忧扩散后，正加强对金融系统中网路安全风险的审查。新加坡金融监管机构敦促银行修补漏洞，南韩各机构开会检视如何回应，澳洲则要求放款机构保持警惕，避免客户因控制薄弱而暴露风险。此番担忧源于该模型仅限量发布，以及有报导称 Anthropic 在发现它能揭露多年未被察觉的安全漏洞后，暂缓了更大规模的推出。

在南韩，金融监管机构上周与产业团体召开紧急会议，讨论与 Mythos 相关的安全问题、现有防护认知，以及维持系统稳定的步骤。另一个委员会和主管机关也开会评估快速变化的网路安全情势，提案包括一套基于 AI 的即时防御系统。在澳洲，Australian Securities & Investments Commission 表示，网路风险正以更大的规模与更高的复杂度上升，并期待持牌机构每天都要主动采取足够的防护措施；同时，新加坡金融管理局正与网路安全机构协调，强化关键基础设施营运者、包括银行的防御。

新加坡网路安全局于 4月15日发布警示，指出类似风险，且美国官员也有所动作：Scott Bessent 和 Jerome Powell 本月与华尔街银行领袖会面，而 Jamie Dimon 表示，AI 虽然能改善防御，却也加剧了网路风险。官员与顾问将 Mythos 描述为一个警讯，显示 AI 可能同时加速漏洞发现与利用，Lee Won-tae 则表示，人在骇客攻击中或许不再掌握主动权。更广泛的含意是，银行与监管机构正把这个模型视为新威胁环境的早期讯号，但由于该模型仅限量发布，而且许多细节仍不确定，因此回应仍然谨慎。

Regulators across Asia are tightening scrutiny of cybersecurity risks in financial systems after concerns spread about Anthropic PBC’s latest AI model, Mythos. Singapore’s financial regulator urged banks to fix vulnerabilities, South Korea’s agencies met to review how to respond, and Australia told lenders to stay vigilant so clients are not exposed by weak controls. The concern is being driven by the model’s limited release and reports that Anthropic withheld a broader launch after finding it could uncover security holes that had gone undetected for years.

In South Korea, the financial regulator held an emergency meeting last week with industry groups to discuss Mythos-related security issues, current knowledge of protections, and steps to preserve system stability. A separate committee and authorities also met to assess rapid cyber changes, with proposals including an AI-based real-time defense system. In Australia, the Australian Securities & Investments Commission said cyber risks are rising in scale and sophistication and expects licensees to be on the front foot every day with adequate safeguards, while Singapore’s Monetary Authority is coordinating with the cyber security agency to strengthen defenses at critical infrastructure operators, including banks.

Singapore’s Cyber Security Agency issued an advisory on April 15 warning of similar risks, and US officials also moved: Scott Bessent and Jerome Powell met Wall Street bank leaders this month, while Jamie Dimon said AI has worsened cyber risk even as it can improve defenses. Officials and consultants framed Mythos as a warning that AI may accelerate both vulnerability discovery and exploitation, with Lee Won-tae saying humans may no longer hold the initiative in hacking. The broader implication is that banks and regulators are treating the model as an early signal of a new threat environment, but the response remains cautious because the model has had only a limited release and many of the details are still uncertain.