UK regulators probe Anthropic model

British financial regulators have opened urgent talks with banks and the National Cyber Security Centre over risks tied to Anthropic’s newest model. (ft.com) The Financial Times reported on April 12 that the Bank of England, the Financial Conduct Authority and HM Treasury were involved in the discussions. Reuters matched that account the same day, citing the Financial Times report. (ft.com, reuters.com) The immediate concern is cyber risk inside finance: whether a stronger model could help expose weaknesses in critical information technology systems used by banks, insurers and exchanges. The National Cyber Security Centre is the United Kingdom agency that advises on cyber threats to government and industry. (reuters.com, ncsc.gov.uk) Large language models are systems trained on vast amounts of text so they can generate code, summarize documents and answer questions in plain language. In finance, that can speed up customer service, compliance work and software development, but it also raises the risk that the same tools could be used to find or exploit software flaws faster. (ncsc.gov.uk, bankofengland.co.uk) The Bank of England and the Prudential Regulation Authority said on April 22, 2024 that they were studying how artificial intelligence could affect their statutory goals and the resilience of regulated firms. The Financial Conduct Authority said separately that it wanted “safe and responsible” use of artificial intelligence in UK financial markets. (bankofengland.co.uk, fca.org.uk) That makes the Anthropic review look less like a one-off meeting and more like a live test of a policy line the regulators had already set out in 2024. Instead of waiting for broad adoption, officials are examining model risk before deployment spreads across a tightly regulated sector. (bankofengland.co.uk, fca.org.uk, ft.com) The National Cyber Security Centre has warned that managers do not need to be technical experts, but they do need enough understanding to question how artificial intelligence tools are being used and secured. Its guidance says fast-moving AI systems can create both benefits and security risks when the technology is adopted before controls catch up. (ncsc.gov.uk) Anthropic did not immediately comment in the Reuters report, and the UK review described by the Financial Times was framed as an assessment rather than an enforcement action. The next step is likely to be more detailed guidance to firms on where these models can be used and what safeguards regulators expect first. (reuters.com, ft.com)