Coinbase cuts fraud response times from days to hours with AI-driven rules engine
Coinbase says it has dramatically sped up its fraud defenses by tightly integrating machine learning models with a high‑speed rules engine, shrinking the time it takes to respond to new scam patterns from several days to just a few hours. The update comes as industry trackers warn that crypto fraud has become a professionally run, AI‑supercharged business.
A two‑track approach: models plus rules
Coinbase describes the redesign as a dual‑track strategy: “models [are] responsible for long‑term defense, rules [are] responsible for rapid response.” Both operate inside a unified framework that lets risk teams capture emerging fraud behaviors with rules, then feed those signals back into ML models to improve long‑term detection.
Automation, faster testing and fewer false positives
The rebuild converted a previously manual, slow rule‑creation workflow into a data‑driven, automated recommendation system. Coinbase restructured data pipelines, automated schema evolution, and introduced notebook‑based analysis tools for its risk teams. Rule backtesting performance has improved by more than 10x, enabling faster trials and quicker deployment of protections as scams evolve in real time.
Machine learning now recommends rule parameters with an explicit goal: “reducing false positive rates while combating fraud and minimizing the impact on normal users.” That balance is critical for a major exchange that handles billions in trading volume and must avoid disrupting legitimate customers.
Building on prior ML work
The upgrade expands on earlier efforts detailed in Coinbase’s blog about advanced machine learning models. The company framed the mission as building “scalable, adaptive, blockchain aware ML systems that enable Coinbase to effectively manage risk for its products” without degrading user experience.
Why the timing matters: fraud has industrialized
The move follows stark industry findings. Blockchain‑intelligence firm TRM Labs estimated global crypto fraud at roughly $35 billion in 2025 and warned that, when underreporting is included, “total annual losses likely exceed USD 200 billion worldwide.” In a separate 2026 crime report, TRM said illicit crypto flows hit a record $158 billion in 2025, noting scam networks are increasingly run like professional businesses and that AI tools are accelerating impersonation and outreach at scale.
Inside Coinbase’s security playbook
Coinbase’s chief information security officer, Philip Martin Lunglhofer, has said the exchange is already using ML to monitor user activity and support chats for signs of scams or account takeovers, and sees expanding “AI‑use cases to detect fraud.” The most recent investment — automated, event‑driven rule generation and a potential “one‑click conversion” of effective rules into model features — is aimed at nudging Coinbase toward a more fully automated risk management system as adversaries themselves wield AI to probe and exploit weaknesses faster.
Further reading
For more context on Coinbase’s security posture, readers can consult Coinbase’s fraud‑focused blog posts on machine learning and compliance, and prior coverage of Coinbase scam activity and broader crypto fraud trends on crypto.news.
Read more AI-generated news on: undefined/news
