AI moves deeper into regulated finance and state power

Reuters reports that Lloyds Banking Group has begun piloting an AI tool through Scottish Widows to help customers with investment guidance, making it the first UK lender to roll out such a product. AI is now being placed closer to consumer-facing financial decision support while the FCA is simultaneously reviewing how AI could reshape market power, supervision and the boundary between guidance and regulated advice.

Reuters also reports that President Donald Trump said Anthropic was “shaping up” and signalled that the company could reverse its blacklisting at the Pentagon. A dispute that began as a military-guardrails confrontation is now turning into a test of how quickly frontier-model providers can move back into classified and defence-adjacent procurement channels.

Regulation

• HM Treasury announced today that the UK will modernise payments regulation during Fintech Week, including work on how payment-services rules should adapt to AI agents and a broader framework for tokenised and stablecoin-based payments. The UK is treating AI-mediated payments as a live regulatory design issue rather than a future hypothetical.

• The Disclosure and Barring Service also published same-date official guidance on using AI tools when contacting DBS. It warns users not to enter sensitive personal data, says AI-generated text must be checked for accuracy, and states that AI tools should not be used to obtain legal advice or make decisions about an application.

• The European Commission, through the AI Office, separately opened seven Digital Europe Programme calls worth €63.2 million to support AI in health, digital skills and online safety. Although this is a funding measure rather than a binding rule, it is still a same-date official intervention that shows where the EU is trying to steer deployable AI capacity and compliance support.

Cases

• Reuters reports that Clarifai said it deleted 3 million OkCupid user photos and facial-recognition models trained on them after the FTC settled with OkCupid over the earlier transfer of photos and demographic data. Data-origin scrutiny is now reaching downstream model artefacts as well as the original dataset.

Academia

• arXiv hosts Making AI Compliance Evidence Machine-Readable, which argues for using OSCAL, the NIST compliance format used in FedRAMP, as a way to structure AI-governance evidence and generate assurance artefacts throughout the model lifecycle. It stands out because it tries to turn AI governance from static paperwork into auditable, interoperable compliance infrastructure.

• arXiv also hosts The Open-Weight Paradox: Why Restricting Access to AI Could Backfire, which argues that effective governance of dual-use AI may require a multilateral institutional architecture analogous in function, though not identical, to nuclear oversight models. Whether or not one accepts the analogy, the paper is useful because it shifts the debate from simple openness-versus-closure claims to institutional design under strategic risk.

Events

• Stanford Law’s AI Initiative lists AI Initiative Speaker Series: Legal’s role on the AI frontier for 23 April 2026 at 12:45pm, featuring Elspeth White of Google DeepMind. It remains one of the sharper near-term events for legal governance because it focuses on the role of lawyers inside frontier-model organisations rather than commenting from outside them.

• The European Commission is also listing Apply AI sectoral deep dive – electronic communications for 30 April 2026, 10:30–11:30 CET. That event is relevant because it links sector deployment, EU digital policy and practical adoption pathways in a regulated infrastructure domain.

Sources: Reuters; HM Treasury; Disclosure and Barring Service; European Commission AI Office; arXiv; Stanford Law; European Commission Events