Footprint is building AI agents to fight a $4.5T financial crime economy — starting with AML and fraud

Footprint

Financial crime moves $4.5 trillion annually, making it roughly the world's fourth-largest economy by Eli Wachs's framing, trailing only the US, China, and Germany. Footprint is building what Wachs calls an AI operating system to shrink that number, starting with the compliance infrastructure used by fintechs, crypto companies, and traditional financial institutions.

The core problem is a structural mismatch: bad actors have been running AI-assisted schemes since GPT-3, while the defense side still relies on roughly 1.5 million human investigators manually reviewing suspicious transactions over days or weeks. Cartels exploit that gap deliberately, using the same volume logic as drug smuggling — create enough shell companies, and enough will slip through.

“Bad actors are using AI far quicker than any bank or financial institution and it's become the fourth largest economy in the world. There's $4,500,000,000,000 moved illicitly each year... The first thing we do is we give them the compute to investigate every case that comes across their desk. And the second is very unique to footprint — we give them the ability to remember every case that they've seen.”

What Footprint actually does

Footprint attacks both layers of that defense. The first is compute: giving compliance teams the ability to investigate every flagged case rather than triaging by hand. The second is memory, which Wachs describes as Footprint's differentiating layer. The system retains every case it has seen, compares new activity against historical vectors, and surfaces patterns before fraud completes.

One live example illustrates how deep the investigation can go. A flagged Russian individual triggered an agent that retrieved Ukrainian newspaper records, read the original Cyrillic text rather than relying on a potentially inaccurate translation, traced the individual's wife to New Jersey, and surfaced a real estate listing in her name. That is the kind of multi-source, multilingual lookup that no human analyst would realistically complete at scale.

Access to hundreds of global databases for fact-checking is what Wachs says separates Footprint from what he calls the previous generation of the space — ML-based detection tools that flagged alerts and then handed the hard interpretive work back to humans. Footprint's architecture puts AI on top of those detection tools and databases, making detection and investigation a single continuous process.

Billing and model strategy

Wachs says Footprint charges by usage, with credits priced according to case complexity. A lightweight fraud screen costs less than a deep investigation requiring thousands of historical embeddings. That structure lets customers scale spend with actual workload rather than pay a flat SaaS fee regardless of volume.

On the model side, Footprint routes across multiple frontier models and handles its own model hosting for banks with restricted vendor policies. One bank recently ingested a 150-page policy document, which Footprint converted into thousands of sub-agents, each operating in its own environment. Wachs attributes compliance AI running roughly two years behind legal AI to precisely that regulatory density and the higher trust bar required to deploy it.

Footprint maintains close relationships with frontier labs, which Wachs says is partly about prompt access — investigations require thinking like a financial criminal, and those prompts look suspicious to content filters without established relationships and white-listing arrangements.

The benchmark problem

Current industry benchmarks are perversely calibrated. AML heads routinely report 95% false positive rates on flagged transactions; some payment systems run as high as 75%. Wachs frames Footprint's ambition as pushing that toward 99.999% precision while simultaneously expanding coverage to every transaction, not just the highest-risk subset. The long-term vision is an agent with memory of every transaction and every account ever opened, able to compare any new activity against that full history in real time.

If AI adds 14% to global GDP by 2030, Wachs argues that either adds $600 billion to the pockets of the world's worst actors, or Footprint and companies like it bring that figure close to zero. Whether the $4.5 trillion figure itself is precisely trackable is harder to say — the nature of illicit flows makes rigorous measurement difficult — but Wachs uses it as the strategic target rather than a verified audit number.