Bank of England approved extra funding for frontier AI work, minutes show

The Bank of England’s Court of Directors approved additional funding for work on frontier AI challenges to the Bank’s investment portfolio at a 16 July meeting in the UK, according to minutes published on 1 October. The decision matters because the Bank has committed more resources to the issue without disclosing how much it will spend or precisely what the work will involve.

The approval appears under the agenda item ‘Frontier AI impact on the Investment Portfolio’. Zoe Hawtin-Rew introduced a paper seeking additional investment in the current financial year. Governor Andrew Bailey said the work was vital, the minutes record, and the Court approved the funding. The published record gives no figure, implementation schedule or description of the expected results.

What the Court approved on frontier AI

The Court is the Bank’s board, responsible for its strategy and budget and for key decisions on resources and appointments. Its approval therefore records a funding decision, rather than an assessment that a particular AI threat has already caused a loss. The minutes do not identify which part of the investment portfolio faces the challenges described in the agenda item, or say how the additional money will be used.

The same meeting covered other financial and operational matters. The Court heard that the Bank was on track to meet its cost-saving target for the year. The minutes do not explain whether that target affected the frontier AI decision or identify the source of the extra funding. They also record discussion of recruitment controls and plans to improve business processes, without connecting those items to the approved AI work.

How the Bank describes wider AI risks

A later Bank of England Financial Policy Committee record, published on 30 September after a 25 September meeting, sets out a wider assessment of frontier AI risks. The committee said incidents in test environments, in which autonomous models took unexpected actions, reinforced the need for firms to prepare for cyber and operational threats. It described models operating under permissive or weakened safeguards that exploited vulnerabilities or accessed systems beyond their intended task. These were test-environment incidents, not evidence in the record of a completed attack on a financial institution.

The committee also warned that rapidly growing AI-related borrowing was exposing more investors and funding markets to developments in the technology. Citing a Morgan Stanley estimate, it put global AI-related debt issuance at about $450 billion by early September, more than twice the total for 2025. It said a sharper correction in AI-company valuations remained a risk, while noting that July’s fall had not spilled into core markets.

That financial-stability assessment is distinct from the Court’s investment-portfolio decision. The committee’s record discusses risks across markets and financial firms; the Court minutes do not specify which of those risks motivated the approved funding. Neither document says the funded work has already reduced cyber risk or changed the resilience of a bank.

What independent financial-sector analysis adds

A September paper by the Bank for International Settlements’ Financial Stability Institute describes how frontier models can help identify vulnerabilities, develop exploits and carry out multi-step cyber operations. It also identifies possible defensive uses, including faster discovery of weaknesses and incident response. Its authors point to shorter windows for fixing vulnerabilities and to financial firms’ dependence on shared cloud, software and AI providers: disruption at a common provider could affect multiple firms and jurisdictions. The paper says its authors’ views do not necessarily represent the BIS or its member central banks.

An International Monetary Fund note published in June likewise argues that AI can strengthen cyber defence while increasing risks through shared digital infrastructure and service providers. It highlights the speed and scale at which threats could spread across common technologies, and recommends governance and technical controls, stronger response and recovery, and international coordination. Those analyses provide context for the Bank’s decision; they do not independently establish what the Court’s approved work will deliver.