Financial institutions racing to embed AI into their compliance operations face a stubborn problem: the technology does not always give the same answer twice. According to Duna, the solution is not to make AI perfect, but to surround it with systems that are.
Compliance, Duna argues, blends fixed rules with case-by-case judgement, and the technology supporting it must handle both. The concern is widespread. The 2026 Global AI in Financial Services Report found that 70% of financial services firms rank model hallucinations and unreliable outputs among their top AI risks, a view shared by the same proportion of regulators. The challenge for compliance leaders is tackling that risk without forcing analysts to check every AI output by hand.
Duna’s answer is to pair AI with a deterministic system, one that applies identical rules to identical information and produces the same result every time. AI, by contrast, is non-deterministic. If a customer breaches a set risk threshold, the deterministic layer triggers extra checks without exception. AI is better deployed where it excels, such as judging the relevance of adverse media.
Duna’s policy engine converts a bank’s policies into code and orchestrates specialised AI agents across onboarding, due diligence, screening, monitoring and perpetual Know Your Customer (KYC). The engine calls on an agent when more evidence is needed, then decides whether requirements are met and what action follows.
A practical example involves value-added tax (VAT) checks. If a registered address differs from the one a customer provided, the engine asks AI to assess the mismatch. The company may have relocated from Belgium to the Netherlands without updating its VAT record, or it may run two headquarters listed in separate sources. Such scenarios are often too rare to merit dedicated rules, so AI evaluates the context and escalates genuinely ambiguous cases to a human.
Before AI outputs can bypass analyst review, Duna says agents must prove themselves, just as analysts do through training and quality assurance. Banks can run agents and analysts on the same cases and compare outcomes, with disagreements revealing either instructions that need refining or cases where analysts themselves diverge. Teams should also rerun identical cases to measure how much results drift when nothing has changed.
Once an agent consistently meets the bank’s standard for a specific task, review can be reduced task by task. At SeQura, Duna performs screening and verification, including adverse media, politically exposed person (PEP) checks, ID verification and watchlist hits, before a case reaches an analyst. Average analyst time per case dropped from 243 minutes to 14.9 minutes, a 16.3-fold improvement.
Duna also stresses designing for reassessment. Because it records the evidence and work behind every case, its system can identify which past cases were affected by an outdated instruction and have AI reassess them using the revised approach.
The takeaway, according to Duna, is that AI need not be deterministic for compliance to be dependable. Trust comes from AI operating inside a system built for consistency.
For more, read the full story here.
Copyright © 2026 FinTech Global
