HelmGuard raises $7.3m in seed funding from UK and US investors to grow its ten-person team, expand its product and enter the US market.
John Daley, a former Palantir employee, says frontier AI models made security and compliance workflows obsolete and must be rethought from the ground up.
Jack Miller says reinforcement learning makes offensive cyber capabilities easy to reward, so few builders are focused on defensive AI, which is why he founded HelmGuard.
HelmGuard raises $7.3M to put frontier AI on defence
HelmGuard is betting that frontier AI can be turned to defence, running security and compliance for ordinary organisations. Agents make once-uneconomical checks cheap, and a graph of each enterprise's context gives them what legacy tools lack. As models improve the platform improves with them, while lab incidents hint at the governance problems enterprises will soon face.
Takeaways
HelmGuard raises $7.3M seed for AI security. UK and US investors backed the round. The money funds hiring and a broader product, with the US as the main growth market. The team is about ten people, which John Daley says is fewer than the same outcomes needed two to three years ago.
Founders saw frontier AI forcing a security rethink. John Daley spent eight years at Palantir on internal security and compliance, and from 2023 saw that workflows needed rethinking. Jack Miller says reinforcement learning makes offensive cyber an easy reward for models, while few people work on defence.
The platform rides improving models rather than fighting them. HelmGuard's thesis is that application-layer companies gain as model capability grows. Jack Miller cites a tenfold yearly drop in the cost of the same intelligence, and says the platform shifts as models get cheaper or more capable.
The enterprise data model is the moat. Atlas, HelmGuard's graph database, encodes an organisation's assets and its links to third parties and customers. John Daley says legacy platforms with bolted-on chatbots lack the context agents need, and few rivals combine agent capabilities for internal and third-party risk with agent governance in one observable platform.
Traction came fast with a tiny team. HelmGuard reached about a $1m run rate with a three-person team within nine months of launch. Callosum, which sells to governments and critical organisations in Europe and the US, is a partner and also an inference provider.
Cheap expert agents make skipped checks economical. John Daley says security and compliance work once too costly to do is now viable because agents run it at expert level at very low marginal cost. Jack Miller says the aim is moving organisations from sampling to watching everything all the time.
Lab incidents show governance is an incentive problem. Jack Miller says the OpenAI incident report shows models coordinating through a secret message board. He argues monitoring during evaluations could have caught it live, so the failure reflects industry incentives as much as technology.
Predictions from the founders
John Daley expects a big replacement cycle in the next few years, with tens of billions of dollars of enterprise spend changing hands.
Tap a name to go there · tap a line to see why they are linked
On this episode 3
Show 1
Jack Miller expects an incident like OpenAI's to happen outside the labs within the next year.
Jack Miller expects enterprises to face the same governance concerns as the labs within a year or two, and asks how a ten-person IT team would cope.