MouseCat's bet: AI agents, not rules, run fraud investigations

What the business is

MouseCat sells AI agents that investigate fraud: they pull a company's data from sources like Databricks or Snowflake, review cases, search databases and APIs, identify fraud rings, and propose backtested rules — running continuously rather than only at transaction time.

Starting capital：Y Combinator W26 accelerator backing; no larger round publicly disclosed as of September 2026.

How it started

Nick Aldridge (CEO), an ex-AWS principal engineer who led Bedrock Knowledge Bases and AgentCore and is one of nine core maintainers of the Model Context Protocol, co-founded MouseCat in early 2026 with Joseph McAllister (CTO), who spent four years building risk infrastructure and fraud models at Coinbase. Friends for six years, they joined YC's W26 batch and launched their platform around the Winter 2026 Demo Day.

What happened

MouseCat positions itself against the incumbent stack — rules, traditional ML models and manual review — which adapt slowly to new fraud trends, only check transactions at point-in-time, and generate false positives as patterns shift. Its agents run continuous post-transaction investigations, integrate with proprietary data on-prem, and turn findings into backtested rules and features. TechCrunch highlighted it as one of the most interesting startups of the W26 batch in March 2026, and the YC company page says it works with Fortune 500 companies and startups; a Launch YC post followed in June 2026.

How it ended up

Still early: an active YC W26 company in New York City with a six-person team, targeting banks, fintechs and marketplaces.

What can be applied

Fraud defenses fail at the investigation bottleneck, not detection: whoever automates the slow manual review and ships rules back into production owns the next layer of risk software.