The AI landscape doesn't move in one direction — it lurches. Some techniques leap from experiment to table stakes in a single quarter; others stall against regulatory walls, technical ceilings, or organisational inertia that no amount of hype can dislodge. Knowing which is which is the hard part. The State of Play cuts through the noise with a rigorously maintained index of AI techniques across every major business domain — classified by maturity, evidenced by real-world adoption, and updated daily so you always know where you stand relative to the field. Stop guessing. Start knowing.
A daily newsletter distilling the past two weeks of movement in a domain or two — delivered to your inbox while the index updates in the background.
Each dot marks the weighted maturity of practices within a domain — hover for a brief summary, click for more detail
AI that screens transactions and entities against anti-money laundering rules, sanctions lists, and watchlists. Includes real-time transaction screening and entity resolution against PEP databases; distinct from transaction fraud detection in Finance which identifies fraudulent payments rather than regulatory violations.
AI-driven financial crime screening has proven its value at leading institutions but remains stuck in a paradox: the tooling works, yet the problem persists. Tier-1 banks routinely achieve 60-90% reductions in false positives through ML-based transaction monitoring, watchlist matching, and entity resolution -- capabilities now backed by GA vendor products, analyst recognition, and explicit regulatory codification in 2026 rulemaking. Agentic AI has emerged as a frontier capability, with production deployments automating 90% of manual investigation effort. The ecosystem is mature enough that the question for large institutions is implementation strategy, not feasibility. Yet adoption remains sharply bifurcated. Current surveys show only 30% of firms use AI for sanctions screening despite it being a high-volume task, and recent regulatory pressure has shifted from volume defense (SAR counts) to outcome measures (coverage, precision, prioritization). The defining tension is no longer whether AI can reduce false positives, but whether systems can be trusted at scale without introducing silent failures (false negatives). Model drift, data quality, and the persistent 95% false positive baseline suggest that efficiency gains at individual institutions do not yet translate into measurable reductions in global illicit financial flows. Until the field can demonstrate systemic impact -- not just operational savings -- advancement beyond good-practice will remain blocked by validation requirements, explainability mandates, and unresolved questions about aggregate effectiveness.
Production deployments now operate at genuine scale and are advancing toward agentic automation. SymphonyAI's new agentic AI agents cut manual investigation effort by 90% with 10x faster case resolution (100 minutes reduced to 10 minutes), achieving 99% false positive reduction through autonomous entity resolution and relationship analysis at a major U.S. bank. Parallel to this tier-1 momentum, adoption remains sharply limited: only 30% of firms currently use AI for sanctions screening despite it being one of the highest-volume compliance tasks, suggesting the technology gap is behavioral, not technical. Silent Eight runs across 150+ regulated markets with 100M+ AML investigations at 98.7% precision. WorkFusion automates 1M+ daily alerts at 10 of the top 20 global banks. Flagright reports 93% false-positive filtration.
Regulatory baseline hardened dramatically in April 2026. FinCEN and federal banking regulators issued proposed rules codifying risk-based AML/CFT standards, explicitly endorsing AI/ML as defensible innovation and acknowledging that imperfect detection is acceptable if risk-calibrated. AMLA's direct supervision model (effective 2028) rejects the 95% false positive baseline as a "control deficiency" rather than an unavoidable cost, demanding contextual risk scoring, integrated signals, and auditable reasoning trails. The EU's 20th Russia sanctions package demonstrates escalation: sectoral bans on crypto platforms, first-ever anti-circumvention tool activation, and $93.3B in detected evasion flows in under one year, signaling regulatory intensification and expanding mandate into digital assets (e.g., first explicit legal mandate for sanctions compliance in stablecoins).
The barriers are structural and evolving. Silent failures (false negatives via fuzzy matching on transliteration variants, model drift as adversaries adapt patterns) have emerged as material risks. Only 11% of practitioners express confidence in their data quality; only 47% report fully connected systems despite 98% AI integration claims. Explainability, validation testing, and vendor governance are now critical blockers even at tier-1 institutions. The OFSI Bank of Scotland £160K penalty exemplified the fragility: a Russian-designated individual's transliterated name variants defeated automated screening despite the sanctioned entity being on the watchlist. The Napier AI Index estimates $183B in annual savings if widespread AI adoption matured, but that conditional remains unproven.
— Jones Day analysis of AMLA enforcement toolkit (2024, direct supervision 2028): three enforcement mechanisms with penalties up to €10M or 10% turnover; signals hardening European regulatory regime with explicit AML/CFT powers.
— Legal analysis of FCA findings on 150+ payment firms: only 44% resolve alerts within 1 working day; recurring failures in fuzzy matching calibration, transliteration handling, and alert governance; regulators expect controls tested in live conditions.
— Record €880M AML fine sought against Nordea (largest in Europe); multiple concurrent enforcement actions signal regulatory escalation and increased scrutiny on AML/sanctions program effectiveness and control governance.
— Canadian OSFI Guideline E-23 (effective May 2027) expands model risk governance to all AI/ML models at regulated institutions including AML systems; explicit scope includes bias checks, explainability, drift monitoring, and autonomous re-parametrization.
— Operational metrics on false-positive costs in sanctions screening ($2-5k daily per mid-market bank) and workforce impact (60%+ attrition); documents alert-to-investigation conversion <1% and cost drivers of unmanaged FP noise.
— Federal Reserve formally extended SR 11-7 (Jan 2026) to explicitly include AI/agentic systems in model risk framework; AML compliance models now treated as high-risk assets requiring rigor governance, transparency, continuous monitoring.
— FCA supervisory thematic review of 150+ firms: exact-name sanctions matching 90% accurate, variant-name matching 75%; alert resolution delays cause bottlenecks; regulators now mandate evidence-based control documentation and continuous tuning.
— Independent Celent analyst brief: agentic AI deployments achieve up to 80% false positive reduction with multi-agent automation of alert-to-SAR workflows including autonomous sanctions screening and web research agents.