The people who defunded terrorists with spreadsheets think they can do the same for rogue AI — and they might be the only ones with a plan that doesn't require every nation to agree first.
The Summary
- Shlomit Wagman, former Israeli financial intelligence chief and current Harvard fellow, proposes using the FATF (Financial Action Task Force) model — the global framework that dismantled ISIS financing — as the blueprint for AI governance
- The framework has backing from unlikely allies: Pope Leo and Anthropic, suggesting cross-ideological viability
- Unlike nuclear arms control analogies, FATF works through risk-based standards and peer pressure, not centralized enforcement or unanimous consent
The Signal
Wagman's core insight is that AI governance debates keep reaching for the wrong historical precedent. Nuclear non-proliferation requires every nation with enrichment capability to opt in. One defector breaks the system. That's a terrible model for technology that can be trained in warehouses from Lagos to Shenzhen.
FATF, by contrast, governs global money laundering through something smarter: standardized risk assessments, transparent reporting requirements, and mutual evaluation. Countries don't need to agree on philosophy. They need to agree on what "know your customer" means and accept that if they don't comply, their banks get frozen out of international finance.
"The framework that choked off ISIS funding didn't require moral consensus — just operational alignment."
The model has three mechanisms that map cleanly to AI risks:
- Risk-based standards that scale from corner shops to global banks (read: from startups to frontier labs)
- Mutual evaluation where peer nations audit compliance (imagine safety red-teams done between countries, not just inside companies)
- Graduated consequences short of sanctions (restricted API access, limited compute exports, excluded from international AI research collaborations)
Anthropic's apparent endorsement signals that at least one frontier lab sees voluntary coordination as preferable to the alternative: legislators who don't know what a parameter is writing rigid capability bans. Pope Leo's backing adds weight from the institutional actor most concerned with AI's moral implications. When the Vatican and a $30B AI company agree on governance architecture, you're probably onto something structural, not ideological.
The Implication
Watch for FATF-style language in upcoming AI policy proposals: "risk-based frameworks," "mutual evaluation," "graduated enforcement." If Wagman's model gains traction, expect the first tests in allied nations with both AI development capacity and financial regulatory sophistication — likely starting with EU members who already operate under FATF compliance.
For AI companies, this means governance won't be a binary "regulated or not" future. It'll be tiered risk categories with corresponding transparency requirements. Build accordingly. For nation-states outside the initial framework, the question becomes: can you afford to be the North Korea of AI, or do you need access to the international model ecosystem badly enough to play by shared rules?