On 26 October the FCA published examples of good and poor AI practice and outlined a trusted framework for agentic finance covering agent identity, delegated authority, accountability and execution. Jessica Rusu, the FCA's chief data officer, was explicit: AI providers must place clear disclaimers on financial content, and autonomous systems need transparent governance chains. This matters immediately to mid-market law firms, insurers, financial advisers and accountancy practices because it signals that the FCA will not accept the current state of play — firms deploying Copilot, Harvey, or bespoke LLM systems for client-facing work or back-office decisions without documented accountability will be seen as non-compliant. Consumer Duty PS22/9 already requires firms to act in consumers' interests; agentic systems that obscure who is responsible for a decision violate that duty by design.

What the FCA framework reveals is that the UK regulatory approach to AI has matured past the 'let's see what happens' phase into active enforcement readiness. We have seen this pattern before: ICO guidance on algorithmic accountability (GDPR compliance), then SRA warnings on document AI in legal services, then PRA SS1/23 on operational resilience including third-party AI risk. The FCA is now saying: we will tell you what good looks like, we will publish what bad looks like, and we will supervise. The ISO 42001 AI management standard is now a de facto compliance baseline. The EU AI Act's influence on UK practice is already visible in this framework — particularly the emphasis on transparency and human oversight of high-risk decisions. Firms that thought they could deploy AI quietly and hope for exemption are miscalculating.

Trovix's approach to this problem differs fundamentally from how most commercial AI products treat agentic systems. Harvey, Luminance, and similar tools are built for efficiency and trained to minimize friction. They are good at what they do. But they typically shift accountability to the end user — 'the AI suggested this, you approved it' — which is legally weak under Consumer Duty and SRA Code because it treats AI as a tool rather than a delegated decision-maker. The FCA framework requires explicit governance: who programmed the agent, what constraints does it operate within, who monitors its outputs, who is liable if it fails. Trovix Audit exists because that governance layer does not come built into most AI products. You need to document the decision chain, version the agent's instructions, log what it decided and why, and prove you caught problems before they became client harm. Products that skip this are not compliant with what the FCA just published.

What you should do now: (1) Audit every agentic AI system your firm currently runs — document what each agent does, who approved its deployment, what it is trained on, and how you monitor its output. If you cannot answer those questions, stop using the system until you can. (2) Do not wait for sector-specific FCA guidance for insurance or legal services — the framework published on 26 October applies across sectors. (3) Use Trovix Watch to track the detailed guidance the FCA will publish over the next 12 months; regulatory detail will emerge and you need early warning. (4) If you are considering new AI agents for client work or regulated decisions, build the compliance case first, not the use case. Accountability architecture is not optional.

Source: Fintech Global