The pilot era is over. Welcome to the accountability era — where good intentions are not evidence, and ‘we think it works’ is not a defence.
Something has quietly shifted in the conversation around AI in financial services. The question firms were asking two years ago – can we use AI? – has been overtaken by a far harder one: can we prove it works the way we say it does?
This shift matters enormously because it changes who’s responsible, what they need to demonstrate, and what happens when they can’t. It is the defining transition of 2026 for every compliance team, every advice director, and every senior manager with their name on an AI deployment.
Over the coming months, we’re going to be taking a deep dive on: AI on Trial: The Burden of Proof — because that transition deserves to be named directly. We’ll examine what accountability in AI actually requires, what the evidence looks like in practice, and what it means for the firms that are doing it right.
The Burden of Proof is on You
In the criminal justice system, the burden of proof sits squarely with the prosecution. In AI governance in financial services, that structure is inverted.
Under Consumer Duty, SMCR, and the evolving expectations of the FCA’s Mills Review, a firm deploying AI shouldn’t wait to be accused. It must be able to demonstrate – proactively, continuously, and at the level of individual interactions – that its AI is performing in customers’ interests. The burden doesn’t shift after something goes wrong. It exists from the moment the model is switched on.
This means the question for every senior manager is not “has anything bad happened?” It is “can I prove, right now, that nothing bad is happening — and that I would know if it were?” Those are very different questions. Most firms can answer the first. Very few can answer the second.
Four Forces Converging in Q2 2026
Four concrete market forces are colliding right now, and together they make the accountability question urgent in a way it simply wasn’t eighteen months ago.
- THE REGULATORY RECKONING
The FCA’s Mills Review, launched in January 2026, is explicitly examining accountability, assurance, and what oversight senior managers must have over AI-driven decisions. Consumer Duty rules are simultaneously under review as AI reshapes advice and intermediation. Comprehensive guidance is expected by the end of the year. Every compliance team in the country is watching — and every Dear CEO letter that arrives in the interim is a reminder that the watchdog is watching back.
- THE DEPLOYMENT GAP
Ninety-nine percent of financial services firms plan to put AI agents into production. Only eleven percent have done so. The largest blockers are governance concerns, cited by nearly half of firms, and data quality concerns. The gap between “we want it” and “we can govern it” is vast — and it is precisely the territory where Aveni operates.
“Only 20% of board-level leaders trust AI agents for financial transactions. Senior managers are personally liable under SMCR for AI decisions made under their oversight. This is not a communications problem. It is a governance infrastructure problem.”
- THE BOARDROOM CONFIDENCE DEFICIT
The trust gap at the top of organisations is structural. When only one in five leaders trusts AI agents for financial transactions — against more than one in three who trust them for data analysis — the issue is not sentiment or familiarity. It is the absence of the infrastructure that would allow a senior manager to look the FCA in the eye and say: we know what our AI is doing, we can evidence it, and we can demonstrate it has not harmed a single customer.
- THE GENERIC AI PROBLEM
Every major platform — Microsoft, Salesforce, the full stack of hyperscalers — is pushing AI deployment faster. What none of them are doing is standing in the room and saying: before we celebrate what this can do, let us show you how you prove it is safe. That space — purposeful, credible, sector-specific accountability infrastructure — is unoccupied. It is Aveni’s ground entirely.
The Charges Brought
Over the next few weeks, we’re going to focus on five real regulatory AI challenges firms are facing today — often without knowing it.
- Deploying AI agents you cannot supervise
SMCR personal accountability; FCA Mills Review oversight requirements
- Giving advice with no retrievable audit trail
Consumer Duty outcome evidencing; FCA suitability record requirements
- Sampling 3% of interactions and calling it oversight
Consumer Duty 100% outcome monitoring expectation; FCA thematic review risk
- Allowing poor guidance quality to scale unchecked
Consumer Duty product and service outcome standards; vulnerable customer obligations
- Building AI on models with no financial services provenance
EU AI Act traceability; FCA model risk expectations; SMCR third-party oversight
These are the governance gaps that exist in most firms deploying AI today. Each one represents a question a regulator could reasonably ask — and that many firms can’t currently answer.
The Goal Is Confidence, Not Caution
This campaign is emphatically, unapologetically pro-AI. The argument is not that AI in financial services is dangerous and should be feared. AI is one of the most powerful forces for good that the industry has ever encountered — the potential to improve outcomes for millions of customers, at a scale and consistency that no human workforce could match, is genuinely extraordinary.
Powerful things deserve proper scrutiny. That is not a counsel of fear. It is the condition that allows powerful things to be trusted — by regulators, by boards, by customers, and by the advisers who deploy them. The firms that embrace accountability infrastructure are not slowing down their AI adoption. They are removing the last remaining obstacle to deploying it at full confidence and full scale.
Aveni is here as the expert witness for the defence. Not the prosecution. Not the regulator. The authority that helps firms build their case, assemble their evidence, and walk into any scrutiny with the confidence of knowing exactly what their AI has done and why.
Follow the series across LinkedIn and our blog through April, May, and June 2026.
