Criminals moved $154B in crypto last year using AI. Chainalysis just gave compliance teams the same tools — a natural language interface to blockchain forensics that its own data has already helped courts trust.
image from Gemini Imagen 4
Chainalysis launched AI agents that translate natural language prompts into blockchain investigation workflows, including fund tracing and audit-ready compliance reports, addressing a critical skills gap in blockchain forensics. The timing aligns with a 162% YoY surge in illicit cryptocurrency receipts to $154 billion in 2025, as AI-enabled criminal networks accelerate fund movement across chains. The agents operate in two modes — deterministic for reproducible, court-admissible outputs and exploratory for open-ended pattern detection across Chainalysis's dataset — reflecting a design philosophy that treats evidentiary standards as non-negotiable regardless of use case.
Blockchain forensics has always required specialists. Reading a chain of transactions, mapping wallet clusters, understanding how mixers and cross-chain bridges obscure funds — it took training and experience that most organizations couldn't afford to maintain at scale. Chainalysis, the blockchain intelligence company whose data has been ruled reliable and admissible in court and has powered more than ten million investigations, thinks AI agents change that calculus. On March 31 at its annual Links conference in New York, Chainalysis announced its first blockchain intelligence agents: software that translates natural language prompts into on-chain investigation workflows, from tracing fund flows to generating audit-ready compliance reports.
The timing isn't accidental. According to Chainalysis's 2026 Crypto Crime Report, illicit cryptocurrency addresses received at least $154 billion in 2025, a 162% year-over-year increase. Stablecoins now account for 84% of all illicit transaction volume. The window for intervention is compressing as AI-enabled criminal networks move and obscure funds across chains in seconds. Chainalysis's pitch is straightforward: if the other side is scaling with AI, the compliance and investigation side needs to do the same.
The agents run in two modes. Deterministic mode produces consistent, reproducible outputs — the same inputs, rules, and data always generate the same outcome — which matters for anything that might end up in a SAR filing or legal proceeding. Exploratory mode is for open-ended signals intelligence: the agent roams across Chainalysis's dataset looking for patterns, and generates an audit trail of what it found. Both modes are designed for the same evidentiary standard. "For critical decisions, the same inputs, rules, and data always produce the same outcome, so automation stays predictable, consistent, and defensible," the company wrote. "In other contexts, you can use agents for signals intelligence and exploration."
That dual-mode architecture is the meaningful design choice, not marketing language. TRM Labs launched a similar product nine days earlier — Co-Case Agent shipped on March 25 — and neither company cites the other. Both appear to have independently concluded that natural-language investigation interfaces are the next productivity surface for compliance teams. Chainalysis CEO Jonathan Levin described the agents as "the evolution of the platform we have built and everything we have learned, billions of screened transactions, over ten million investigations, more than a decade of blockchain intelligence, that will work alongside your team." The agents begin rolling out over summer 2026, starting with investigations and compliance workflows.
The $154 billion figure is a lower-bound estimate — Chainalysis identifies illicit addresses it can confirm, not the full universe. But even accepting that caveat, the scale of what's flowing through blockchain networks is forcing a rethink of how compliance operates. Impersonation scams grew 1,400% in 2025. Ransomware attacks increased 50%, though payments declined 8% as organizations improved backup and response. North Korea's hackers stole $2 billion in crypto in 2025, including the $1.5 billion Bybit exploit — the largest digital heist in crypto history.
The governance question the announcement doesn't fully answer is who decides what "deterministic" means in practice. Chainalysis says its agents are "designed for regulated, high-stakes decisions" and that "humans decide what can be automated and at what level of independence." But in a world where regulators are still figuring out what AI-assisted decision-making requires legally, that promise is a policy position, not a technical guarantee. The audit trail is real. Whether it's sufficient for a given regulator is not something Chainalysis can resolve by announcement.
What the agents actually are, technically: a natural language interface on top of Chainalysis's existing dataset and investigation workflows, with an agentic layer that can chain actions — enrich alerts, build dashboards, identify time-windowed transactions, orchestrate multiple agents in parallel. During early development, Chainalysis used them internally for workflows that "used to take days compressed into minutes, with full audit trails." That's the productivity claim. Whether that compression holds in production, with real compliance teams and real adversaries, is the open question.
The democratization framing — putting blockchain forensics into the hands of non-specialists — is plausible. The dataset has been ruled reliable and admissible in court, which matters for the legal defensibility of any output. But specialized skill and reliable data are different things. A tool that surfaces the right information doesn't automatically produce the right analysis. Chainalysis is betting that the agent layer bridges that gap. That's an empirical claim, not a technical one. The answer arrives when the summer rollout starts and investigators actually use it.
Story entered the newsroom
Assigned to reporter
Research completed — 6 sources registered. Chainalysis launched blockchain intelligence agents at Links 2026 (March 31), summer rollout. Agents built on 10M+ investigations and billions of scre
Draft (716 words)
Approved for publication
Published (735 words)
Get the best frontier systems analysis delivered weekly. No spam, no fluff.
Agentics · 2h 30m ago · 4 min read
Agentics · 2h 48m ago · 4 min read
