Most AI agent startups start with email and calendars. Lumbra started on classified networks. The Washington company announced a $25 million Series A on November 10, co-led by Point72 Ventures and Ansa with Costanoa returning, and it already has agents in production on top-secret networks across the Pentagon and the US intelligence community.
Lumbra was founded in 2025 by Aaron Brown, a former CIA officer, and Michael Hiebert, an MIT-trained engineer and former special operations infantry officer. The team includes people from Palantir, Anduril, Northrop Grumman, JSOC, NSA and CIA.
An operating system for agents
Lumbra calls its product an operating system for AI agents in high-consequence environments. The harness, Nebula, is built to make agents reliable and auditable for real analytical work. Its StarChart component is a knowledge graph where agents record findings as they go, alongside sources and sensitivity labels, so institutional knowledge builds up instead of evaporating when an analyst leaves.
The hard part in intelligence work isn’t getting a model to write an answer. It’s proving where the answer came from, keeping classified data where it is and making sure an agent working across several tools doesn’t leak information between classification levels. Lumbra’s pitch is audit trails and controls that turn AI output into something a decision-maker can sign off on.
Next stop, banks and hospitals
The new money goes into engineers who will bring the same controlled agents to finance, healthcare, law and other regulated sectors. That’s a sensible path. A system that satisfies the intelligence community’s rules on data movement and provenance should clear a compliance team at a bank.
The risk is the one every govtech startup carries: procurement cycles are slow and one stalled pilot can freeze expansion. Moving into commercial markets now spreads that risk while the government business keeps growing.