Morgan Stanley has announced that external AI agents can now access ShareWorks and Equity Edge directly—bypassing human advisers entirely. This is not a small move. For UK financial services firms watching from across the Atlantic, it signals a hard competitive choice: open your core systems to third-party AI tools or risk client friction. But there is a critical difference between American speed and British regulation. The FCA's Consumer Duty (PS22/9) requires firms to act in clients' interests and demonstrate that they have done so. Simply opening a platform to an external AI agent—however well-intentioned—does not automatically satisfy that duty. UK regulated firms cannot follow Morgan Stanley's playbook without asking themselves a harder question first: who is liable when the agent fails?

This story is part of a larger pattern. AI vendors are pivoting from "co-pilots for humans" (think Microsoft Copilot, Legora for legal research) to autonomous agents that make decisions with minimal human oversight. The industry narrative is that this is inevitable, that humans are the bottleneck, that agents are faster and cheaper. There is truth in that. But the pattern also reveals something uncomfortable: most organisations implementing agentic AI have not solved the governance problem. They have simply pushed the liability question downstream. Harvey and other AI legal platforms have built human-in-the-loop workflows because their users—law firms—have told them that FCA Consumer Duty, SRA Code of Conduct, and the FRC's ISA UK all demand it. Morgan Stanley's move is less cautious. It assumes that having audit trails and disclaimers is enough. For UK financial services and professional services firms operating under tighter regulatory scrutiny, that assumption is dangerous.

Our view at Trovix is direct: agentic AI in regulated industries must be designed for compliance first, not speed first. When we built Trovix Reach for client-facing interactions and Trovix Audit for governance oversight, we started with the constraint, not the capability. The constraint is that a UK regulated firm cannot delegate judgment to a black box and walk away. Audit logs, bias testing, and model cards are not optional add-ons—they are the product. Morgan Stanley's approach assumes that large language models trained on market data will make better decisions than humans at scale. They may be right for equity administration. But for regulated advice, for risk decisions, for anything touching Consumer Duty, the assumption fails. We have seen firms try to retrofit governance onto agents built for speed. It is expensive and it often does not work because the agent was never trained to be auditable. The alternative—building for auditability from the start—takes longer but it actually works.

If you are a mid-market wealth manager, insurer, law firm or accountancy practice, here is what to do now. One: audit your current AI tooling. Are your document processors, research assistants, and client-facing chatbots genuinely auditable under ICO UK GDPR and your sector regulator's expectations? Two: when you evaluate new agentic AI (whether from vendors or built in-house), demand an auditability specification before you demand a capability spec. Three: do not assume that following Morgan Stanley is safe. Morgan Stanley has regulatory capital, legal resources, and an army of risk officers. You do not. If your chosen AI agent makes a decision that breaches Consumer Duty or creates an insurance claim, you need to be able to prove that you acted in the client's interest and that you had controls in place. Morgan Stanley can afford to move fast and manage the regulatory conversation later. You cannot.

Source: CNBC