UK regulators urgently assessing Anthropic model

British financial regulators have opened urgent talks with banks and the National Cyber Security Centre over cyber risks tied to Anthropic’s newest artificial intelligence model. (reuters.com) Reuters reported on April 12 that the meetings involve the government’s cyber agency and major banks, and that briefings for banks and exchanges are expected soon. Bloomberg separately reported on April 11 that the Bank of England planned discussions with financial firms about Anthropic’s new model. (reuters.com, bloomberg.com) The model at the center of the review is Anthropic’s latest Claude release. Anthropic says Claude Opus 4.6 is its strongest shipped model and highlights software engineering, tool use, search and finance as core use cases. (anthropic.com, anthropic.com) Large language models are systems that predict and generate text, code and actions from patterns in training data. In finance, banks already use artificial intelligence for fraud detection, customer support, document review and coding, according to a 2024 joint survey by the Bank of England and the Financial Conduct Authority. (bankofengland.co.uk) That survey said 75% of firms were already using artificial intelligence, up from 58% in 2022, and 10% had no plans to use it. It also identified data quality, third-party dependence and cyber security as central risks for firms adopting these tools. (bankofengland.co.uk) The cyber concern is straightforward: more capable models can help defenders find flaws faster, but they can also help attackers identify weak points in critical systems. The National Cyber Security Centre said in its 2025 assessment that artificial intelligence will increase both the volume and impact of cyber intrusion activity through 2027. (ncsc.gov.uk) The National Cyber Security Centre is part of Government Communications Headquarters, Britain’s signals intelligence agency, and advises companies and public bodies on digital threats. Its involvement signals that officials are treating the issue as more than a routine model-governance review. (ncsc.gov.uk) Anthropic has publicly framed its newest Claude models as strong at coding and “agentic” work, meaning systems can carry out multi-step tasks with limited human prompting. Those are the same capabilities that make banks interested in automation and regulators wary of mistakes or misuse inside sensitive networks. (anthropic.com, anthropic.com) The UK review also fits a broader pattern. The Financial Times stream on Anthropic reported on April 8 that United States bank chiefs had been called to discuss cyber risks linked to Anthropic’s model, showing that concern is not confined to Britain. (ft.com) The immediate question for British lenders is whether the model can be used inside core systems without creating new paths for attack or failure. The answer is likely to shape the guidance banks and exchanges receive in the next round of regulator briefings. (reuters.com, bloomberg.com)