SuperML
← All issues

CommBank's Fraud Agent Now Writes Its Own Detection Rules — The Architecture Shift Behind a 20% Drop in Fraud Losses

CommBank's agentic fraud AI now writes 75% of its own card detection rules — and delivered a 20%+ reduction in fraud losses — but the architecture behind human-in-the-loop rule generation at 80M daily signals is what every fraud AI team should be studying.

Hi there,

CommBank just published the production numbers on an agentic AI system that doesn't just detect fraud — it proposes new detection rules when it finds new patterns, then routes them to human analysts for approval. The result: 75% of card fraud rules now have the agent's fingerprints on them, and fraud losses dropped 20%+ in the first half of FY2026. The governance architecture enabling this is the real story.

🔥 Featured Post

CommBank's Fraud Agent Now Writes Its Own Detection Rules — The Architecture Shift Behind a 20% Drop in Fraud Losses

  • Agentic fraud AI monitoring 80M+ daily signals and autonomously proposing new detection rules in real time
  • 75% of CommBank's card fraud rules now generated or updated by the agent — with human approval gates
  • 20%+ fraud loss reduction in first half of FY2026 on a $1B annual fraud safeguarding investment
  • Built on Snowflake data cloud + cloud-native core banking — the data architecture is load-bearing
  • The human-in-the-loop review loop is not optional governance theater; it's the failure mode prevention layer

Read the full post →

📚 In Case You Missed It

When Your AI Vendor Becomes Your Systems Integrator: The Enterprise Architecture Reckoning Behind the OpenAI-Anthropic PE Playbook — OpenAI's $10B 'Deployment Company' and Anthropic's $1.5B Blackstone-Goldman venture both launched May 4 with the same playbook — embed engineers, redesign workflows, lock in the model — and neither enterprise AI governance framework was designed for a world where your model vendor IS your systems integrator.

Cerebras Files for $26.6B IPO With OpenAI as 86% of the Backlog: The Wafer-Scale Tier Just Became an Architecture Decision — Cerebras filed for a $3.5B IPO at $26.6B valuation on May 4, with OpenAI's 750 MW Master Relationship Agreement and an $1B circular loan baked into the S-1 — making wafer-scale inference a real architectural tier and a real concentration risk that enterprise AI teams now have to model into their LLM gateway, latency budget, and vendor strategy.

The NL-2-SQL Agent Trap: Why LLMs Need an Ontology Layer to Stop Hallucinating Your Database — Google's BigQuery + Gemini NL2SQL pipeline exposes a dirty secret: LLMs alone can't reliably generate SQL over enterprise schemas — they need an ontology layer that maps business language to tables and columns, or you get syntactically valid but semantically wrong queries at scale.

More posts dropping every day. Stay curious.

— Bhanu @ superml.dev