英国政府官员正与Anthropic进行磋商，英国银行及其他企业正争取更快取得Claude Mythos，以利用其识别网络弱点能力。Anthropic迄今仅向约40家机构推出该模型，参与者主要为美国企业，包含Amazon、Microsoft、JPMorgan Chase及Morgan Stanley。Anthropic向《金融时报》表示未承诺英国广泛部署的时间表，而政府部门正在讨论向关键国家机构扩展存取。

此种紧迫性同时结合了安全与风险考量。英国一位高阶主管称，与已获得该模型的美国公司正在就Mythos揭示的具体漏洞展开讨论，并与Microsoft协商先行建立修补机制，以便在完整接入前加固系统。据悉，JPMorgan Chase执行长Jamie Dimon曾提醒英国银行界须谨慎部署，且应在政府协调下进行。监管者与央行领袖提出同样警告，德国联邦银行行长Joachim Nagel亦呼吁各机构平等获取，避免能力不对称与滥用，并将其描述为“双刃剑”。

Anthropic本月初推出Mythos时宣称其可比人类更快发现网络缺陷，这引发全球对其被恶意利用制造攻击脚本的忧虑；加之本周传出有用户可能透过第三方未授权存取的报道后，疑虑更升。英国是唯一除美国外取得试用权的政府，透过AI Security Institute测试，后者指出Mythos可利用专家需数日才能完成的漏洞。科技大臣Liz Kendall与安全大臣Dan Jarvis发出联合信件警惕AI网络能力加速增长，并敦促企业强化防御；在与英国财长Rachel Reeves的圆桌会晤中，银行、保险与交易所等亦达成以AI缩减攻击面、提升威胁探测与自动化缓解措施的共识，政府则称将持续与业界密切合作维护基础设施韧性。

UK officials are in active talks with Anthropic, and UK lenders and other firms are pressing for faster access to Claude Mythos because it can identify cyber vulnerabilities. Anthropic has so far rolled out the model only to about 40 organisations, mostly U.S.-based, including Amazon, Microsoft, JPMorgan Chase and Morgan Stanley. Anthropic told the FT it has not committed to a UK-wide timeline, while ministers discuss expanding access for critical national organisations.

The urgency reflects both security needs and risk concerns. A UK executive said discussions are under way with U.S. companies on specific vulnerabilities exposed by Mythos, and firms are negotiating with Microsoft for patches so they can harden systems before full access. Jamie Dimon of JPMorgan Chase has reportedly warned UK bankers to deploy the tool carefully and with government coordination. Regulators and central bank figures echoed this, including Bundesbank chief Joachim Nagel, who called for broad institutional access to prevent capability gaps and misuse, calling the model a “double-edged sword”.

Anthropic unveiled Mythos earlier this month, claiming it can detect cyber flaws faster than humans, which has raised fears of exploitation and malicious use, especially after reports this week of possible unauthorised access via third parties. The UK is the only non-US government to have preview access through the AI Security Institute, which said Mythos can exploit weaknesses that experts would need days to uncover. UK ministers Liz Kendall and Dan Jarvis warned that AI cyber capability is accelerating, and called on businesses to strengthen defences. In a roundtable with UK chancellor Rachel Reeves, banks, insurers and exchanges agreed to use AI to shrink attack surfaces, improve threat detection, and automate mitigation and response while officials said they would continue close work with industry to protect infrastructure resilience.