What happens when software stops assisting analysts and starts doing their jobs? In blockchain intelligence, that future may already be arriving.

“We want to automate the tasks of our customers as much as possible,” Emmanuel Marot, vice president of products at Chainalysis, told PYMNTS. “And at the same time, make it easier for them to achieve some of those tasks and help them fulfill their mission.”

Marot pointed to a digital asset landscape where financial investigation is becoming a machine-speed problem, one that humans alone can’t keep up with.

“The amount of content created for scamming people is absolutely through the roof,” he said. “Our customers are swamped.”

Chainalysis’ newly introduced “blockchain intelligence agents,” unveiled earlier this March, signal a shift in how financial crime investigations and compliance workflows may be conducted. Rather than simply visualizing transaction flows, these systems aim to actively participate in the investigative process.

The push reflects a broader truth across finance: as fraudsters weaponize artificial intelligence (AI) to scale scams, institutions are racing to deploy systems that can match that speed without sacrificing control or accountability.

But if the direction of travel is clear, the constraint is even clearer. In high-stakes domains like compliance, law enforcement and financial crime, AI doesn’t just have to be useful. It has to be right, and provably so.

“It’s very easy to duct tape a large language model on top of an existing product,” Marot said. “It’s fancy, but not necessarily reliable—and not very useful, especially for our customers.”

See also: Chainalysis Launches AI Agents for Crypto Crime Investigations

Software That Does the Work

For more than a decade, blockchain intelligence has been a power-user game. Platforms allowed investigators to map flows of funds across wallets, identify patterns and build cases manually. The tools were powerful but also complex, requiring specialized training and time.

Agentic systems, like the one Chainalysis has built, aim to compress that process. Instead of asking an analyst to manually trace a transaction, an AI agent can take a prompt like, “Where did this money come from? Is it suspicious? Where did it go next?” and begin assembling the answer on its own.

“These are end-to-end mini investigations,” Marot said, while being careful to frame the technology as assistive, not autonomous.

“AI should be governed by humans, not the other way around,” he added, noting that Chainalysis is positioning its own agents as “glass box” systems: fast and automated, but fully auditable. “Our users remain in control all the time.”

After all, while a hallucinated answer in a chatbot is an inconvenience, a hallucinated conclusion in a financial investigation could mean a missed crime, a false accusation or a regulatory failure.

“We care about having data the agents can rely on,” Marot said, emphasizing that outputs must be “as good as if the human beings were doing the work.”

“A beautiful black box would be extremely dangerous,” he added.

At the same time, Chainalysis has already experimented with offensive uses of AI, including bots that engage scammers directly to extract intelligence. In some cases, those interactions yield not just crypto wallet addresses, but traditional financial details like bank accounts—blurring the line between on-chain and off-chain investigations.

Winning the Arms Race

The timing of the launch is not accidental. As AI becomes more powerful, it is also being weaponized by criminals and creating an unavoidable surge in the scale and sophistication of attacks.

As a result, compliance teams increasingly face a familiar pattern: large volumes of alerts, most of which are benign, each requiring time-consuming investigation. Agentic systems are designed to absorb that load. They can triage alerts, gather relevant context, apply predefined rules or playbooks and surface conclusions faster. Even partial automation—reducing investigation time from ten minutes to two—can translate into significant gains at scale.

“You and I won’t be smarter in six months. But the models will be better,” Marot said.

Underlying all of this is a rapidly evolving blockchain landscape. The early days of focusing on Bitcoin and Ethereum are long gone; today’s environment spans a fragmented ecosystem of networks, each with its own quirks.

“Nobody can afford to say, ‘I only care about Bitcoin or Ethereum,’” Marot said.

New transaction types, such as multi-recipient payments embedded in a single transfer, introduce additional layers of complexity. Meanwhile, the rise of smart contracts has accelerated both innovation and risk. For Chainalysis, the stakes are particularly high. As blockchain adoption expands—from cross-border payments to tokenized assets—the need for trustworthy, scalable oversight will only grow.

“It sure looks bright,” Marot said of the future. “There’s a real-world usage, and a need to make sure that the money goes to the right place.”

In that future, the role of the human analyst may shift from operator to orchestrator, guiding systems that can act at machine speed but still answer to human judgment.