Every year, ACAMS The Assembly Las Vegas gives the anti-financial crime community a chance to step back from the daily grind of alerts, cases, and exams and ask a bigger question: are we actually getting better at this? This year’s theme, “Insight vs. Noise,” put that question front and center. Exorbitant amounts of information don’t automatically translate into insight — the work is homing in on the data that can actually move a compliance program forward. Across three days at the ARIA, 140-plus speakers and roughly 78 sessions wrestled with how to find it.

Our team participated at the cross-section of those conversations, and a few threads stood out—both because they matched what we heard directly from attendees, and because they echo exactly where we’ve been investing as we continue to evolve products.

The regulatory reset is real, but it isn’t “do less”

The shifting regulatory landscape was a major undercurrent, and the picture that emerged was more nuanced than a simple loosening or tightening. The current U.S. administration’s stated priorities, as outlined in the Regulatory Outlook Masterclass, include fighting cartels, reducing regulatory burden, ending unlawful debanking, and promoting technology. In addition, examiners are looking to tie matters requiring attention to substantive violations and material financial risk—though, there was skepticism over whether they would actually follow the new rule. Even if examiners do stop citing process gaps directly, the underlying processes are still what ensures case quality and completeness — which is to say, still what prevents the material financial risk examiners care about in the first place.

Given this, and the operational demands required to make sweeping changes in a short time, many firms are likely to maintain their current approach. That same caution showed up wherever a new rule asked institutions to move quickly. The proposed anti-money laundering / countering the financing of terrorism (AML/CFT) programs rule is one example: it emphasizes risk assessments and, in practice, may add burden without offering real reform. Notably, speakers were blunt about “snapback” risk — the political winds shift, and the cuts a program makes today can look like the penalty it earns in two years.

That hesitation about big structural change hasn’t stopped new, narrower requirements from landing in the meantime. Recent Geographic Targeting Orders have added requirements with little lead time, including new data fields like zip codes and counties, and the Corporate Transparency Act remains suspended rather than resolved. The common thread: less certainty, not less scrutiny.

That shows up clearly in how the conversation around SARs has changed — the ask isn’t for fewer; it’s for better. The DOJ’s new National Fraud Enforcement Division—the department’s first new division in two decades and now roughly 500 people deep working alongside 70-plus federal Offices of Inspector General—is asking institutions for more detailed SARs and more proactive communication, not a lighter touch. During the panel, Investigator’s Playbook, panelists, including Jonathan Weiner, Global Head of Financial Crime Compliance at eClerx, described what law enforcement actually wants: state the suspicious activity up front, keep the narrative chronological, include communications between the bank and the customer, and link the activity to known typologies. Lengthy narratives are fine if they are substantive and well organized, but poorly formatted ones overwhelm reviewers — as Jonathan shared, “Law enforcement isn’t asking for less detail — they’re asking for a narrative they can actually follow.”

AI is moving from pilot to practice, with governance as the gating factor

The second major theme, “From Experimentation to Implementation: AI and Digital Assets,” captured something we’ve seen first-hand with our own clients this year: the AI conversation in financial crime compliance has changed.

The AI conversation has shifted from “can it touch AML?” to “how do we control it?” One session on AI governance cited false positive reductions of roughly 18–35% alongside more true positives surfacing, which means more findings for investigators to handle. Speakers stressed governance: drift monitoring, bias testing, backup controls if AI fails, and kill switches for agentic systems, since regulatory guidance on agentic AI is still thin. One practical suggestion was to inject deliberate errors to test whether human reviewers actually catch bad AI output, rather than treating “human-in-the-loop” as a sufficient control. The same session expected widespread adoption of AI agents in AML to take longer than a year.

That pressure wasn’t abstract to the people in the room. “Audience polls showed how much tension investigators are feeling around AI right now,” Jonathan said. “Managers are pushing teams to adopt it, and at the same time investigators are genuinely unsure whether it’s a tool for them or a threat to their role — and whether they can trust what it gives them. AML leaders need to settle that question explicitly: the human stays in charge, the tool is built to serve the investigator rather than replace their judgment, and their signature on an output should mean they’ve actually confirmed it’s theirs — not that they clicked through it.”

That’s a hard standard for most legacy alerting workflows to meet, because they were built to flag activity, not to reduce false positives while still explaining every decision in a way a reviewer — and eventually an examiner — can stand behind. It’s also the specific gap we’ve been building against with Compliance Manager AI, our platform for applying AI across the AML/KYC lifecycle It automates the repetitive triage work, but keeps a reviewer positioned at the decision points that matter and attaches rationale to every output, so “the model said so” is never the answer when someone asks why an alert was closed.

Fraud and typology convergence keep accelerating

“Fraud: The No. 1 Threat” wasn’t just a session track — it was the dominant operational concern. Fraud, money laundering, and sanctions evasion are converging faster than most institutions’ org charts or systems were built to handle. The money mule session was a good illustration—it examined how gig-economy culture has normalized remote, informal work to the point that many mules genuinely don’t realize they’re breaking the law, and because their accounts look legitimate on paper, detection gets harder, not easier. Speakers pointed to batch transactions and payment-flow patterns as better tells than individual account behavior, and flagged that criminal networks are now using AI themselves to scale recruitment. The sanctions conversation made a parallel point from a different angle: money laundering, sanctions evasion, and export violations share enablers and middlemen, but most institutions still run them as separate programs with separate teams. Screening against a names list catches the obvious case and misses the network behind it.

The fix that kept surfacing — in the sanctions sessions and in the cross-sector information-sharing conversations alike — wasn’t a better point solution for each typology. It was shared visibility: one investigator able to see fraud, AML, and sanctions signal on the same case, instead of three separate alerts in three separate systems that never talk to each other. That’s the problem ELM Orchestrator—our platform designed to serve as the control layer for client lifecycle execution—solves. By providing a single case view that pulls workflows together, it prevents the need for the person investigating to reconstruct the connection manually, three logins at a time.

The bottleneck is still data quality

Underneath every one of those themes was a quieter, less glamorous one: none of it works without clean, current, well-governed data. A regulator asking for better SAR quality, an AI model that needs explainable inputs, a fraud investigator trying to connect typologies — all of it depends on KYC files and data sources that are actually current and verified, not just checked off at onboarding.

A session on biometric authentication made the point sharply: liveness detection alone can’t catch deepfakes anymore, and the fix isn’t a better scan, it’s “perpetual KYC” built on behavioral and transaction signals that keep updating after onboarding ends. A separate session on moving institutions from fixed-schedule periodic reviews to event-driven ones named fragmented data as the single biggest roadblock to making that shift — not a lack of will, but data spread across systems that doesn’t reconcile cleanly enough to trust. Both land on the same conclusion: current, verified data is the actual constraint, whatever the surface-level conversation is about.

That’s the specific gap our EDOS and ELM KYC solutions are built to close — EDOS by continuously sourcing and verifying data from external sources rather than treating it as a one-time check, and ELM KYC by making sure each case gets the right requirements the first time, with the evidence to prove it, instead of discovering the gap during a review.

Where this leaves us

If there’s one takeaway to carry out of Las Vegas, it’s that the institutions making real progress aren’t the ones running the most AI pilots. They’re the ones who’ve connected the dots between regulatory expectation, data quality, and the person actually doing the investigation — while keeping their programs steady through a political climate that keeps shifting underneath them. That’s the thread we’ll keep pulling on with clients in the months ahead, and we’d welcome the chance to continue the conversation with anyone we met, or missed, at the show.