Engineering deep-dives, whitepapers, and case analyses. What we've learned about the shape of modern fraud, and how we think immune systems should be built. Honest writing, no marketing dressed up as research.
Labels are scarce, attackers adapt fast, and supervised models lag the threat by months. The math behind why an unsupervised approach beats a supervised one for fraud detection in 2026, with code samples and benchmark results from our pilot deployments.
Read the articleA teardown of an account-takeover ring we observed across three pilot deployments. Patterns, signals, and what worked.
Read →Building a unified scoring layer that watches both EVM events and SEPA flows in the same context window.
Read →New pieces published every 3 to 4 weeks. Subscribe below to get them by email, or check back here.
How we tell autonomous agents apart from human users without breaking either experience. Detection signals, allowlist patterns, and the trade-offs we made.
What "explainable AI" actually means under PSD2, GDPR, and the AI Act. How we structure decision logs so your auditor signs off without a fight.
An analysis of 47 wallet-drainer campaigns and their shared infrastructure. Counterparty patterns, signing-permission red flags, and detection heuristics.
The architecture decisions behind our scoring pipeline. Why we don't use deep models in the hot path, and how we handle context lookups without breaking the latency budget.
Our deepest research comes from the partners we work with. Apply for the cohort and your fraud problem might end up in a field note.