Clarent · Our story
AI agents are making consequential decisions in UK-regulated financial services right now, without the governance infrastructure to prove those decisions were lawful, accountable, or fair. Clarent exists to close that gap.
01
Multi-agent systems span Microsoft Copilot Studio, Google Vertex AI, ServiceNow, and bespoke in-house code. Each emits its own logs in its own schema. No tool stitches them into a single chain of reasoning.
02
SMCR requires a named accountable individual for every significant function. For AI-driven decisions, that link doesn't exist today. It's assumed, not evidenced.
03
Existing compliance tools monitor human behaviour. They have no concept of agent delegation, sub-agent calls, tool invocations, or autonomous escalation chains. The category doesn't exist yet.
I studied economics at the University of Edinburgh, a discipline built on understanding why systems fail, how incentives distort outcomes, and where accountability breaks down. That analytical foundation shapes how I think about AI governance. Clarent isn't primarily a tooling story. It's an accountability problem: autonomous systems are making consequential decisions, and the infrastructure to evidence who was responsible, and whether those decisions were sound, doesn't yet exist at scale.
After Edinburgh I moved into data engineering in financial services, working specifically within Risk. Being embedded in a Risk function is a particular vantage point. Every system built there has to be auditable, every decision traceable, every control defensible to a regulator. I worked closely on production generative AI projects within Risk. Not prototypes, not internal demos, but systems making real decisions in environments where the FCA could ask about any output at any time. That experience is the foundation Clarent is built on.
"Shipping generative AI in a Risk function means living with regulatory accountability every day. That's not a constraint I came to later. It's where I started."
Alongside that hands-on production work, I was involved in discussions about what the industry was calling 'super agents'. Autonomous systems spanning Microsoft Copilot Studio, Google Vertex AI, ServiceNow, and bespoke in-house tooling, all converging into a single interface. These were not whiteboard exercises. They were strategic programmes, moving toward production at one of the UK's largest FinTechs.
In every one of those conversations, the same structural gap surfaced. Each platform emitted its own telemetry in its own schema. There was no unified way to stitch them into a coherent chain of reasoning, and no way to map each decision node to the Senior Manager accountable for it. The Consumer Duty obligation and the SMCR personal accountability requirement were already in force, but the infrastructure to evidence compliance simply didn't exist.
"There was no unified place where all agent decisions, sub-agent delegations, tool calls, and chains of reasoning lived in a single auditable record, with regulatory rules built in from the start."
I'm not approaching this problem from the outside. I've been a practitioner inside it: in Risk, on production AI deployments, in the strategic conversations about where multi-agent systems were heading. I saw what was missing before it had a name. That's why I built Clarent.
Clarent is not a general observability tool. It is a purpose-built regulatory compliance layer for AI agents operating in UK-regulated financial services.
A lightweight SDK wraps any Python or TypeScript agent with almost zero latency. Data is PII-scrubbed in-process before it ever leaves your environment, then shipped via OpenTelemetry Protocol to Clarent's infrastructure.
Individual spans from every agent, sub-agent, tool call, and human escalation are stitched into an ordered, attributed decision chain, regardless of which framework or platform each component ran on.
Each chain is evaluated against Consumer Duty (PRIN 2A), SMCR accountability rules, SS1/23 model risk principles, and the DUA Act 2025. The output is an HMAC-signed verdict record that cannot be altered after write.
Consumer Duty has been in force since July 2023. The FCA's April 2025 AI Update confirmed it applies directly to AI-enabled customer journeys. Firms are not waiting for new guidance to be in scope. They are already in scope.
The FCA's Mills Review, launched in January 2026, is expected to produce specific agentic AI guidance in Q3 2026. The EU AI Act becomes fully applicable in August 2026, introducing a 10-year audit trail requirement. A compliance deadline is approaching that is both visible and firm.
Firms that build governance infrastructure now will enter Q3 2026 ready. Firms that don't will be reacting, under regulatory scrutiny, without the institutional knowledge, and without an audit trail for decisions that have already been made.
Regulatory milestones
Consumer Duty in force
Creates the obligation.
FCA AI Update
AI agents in scope.
Mills Review
Urgency intensifies.
FCA guidance expected
Hard deadline.
Verdict records are append-only and cryptographically signed. Governance infrastructure must be tamper-evident to be credible.
Consumer Duty, SMCR, SS1/23, and the DUA Act are not add-ons. They are the structural skeleton of the policy engine. UK regulatory specificity is the moat.
The SDK adds sub-50ms overhead and scrubs PII in-process. Governance should not require rearchitecting your agents.
Every root-level decision chain maps to a named SMF. Compliance is not a system property. It is a human responsibility.
Clarent exists so that when the FCA asks how your agents made their decisions, you can produce a complete, signed, regulator-ready audit trail. Not a week-long forensic exercise.
No self-serve. A Clarent solutions engineer will respond within one business day.