Why Emergent Ventures backed Hypereum.
Non-dilutive capital, regulated industries, and the case for backing infrastructure before revenue.

---

In April 2026, Hypereum was selected for Emergent Ventures, the Mercatus Center’s programme at George Mason University. The programme’s stated mandate is to identify and support underrated talent and ideas, including those operating outside the conventional venture pipeline. The grant is non-dilutive. We retain full ownership of the company and of the systems we have built.

The fact of the selection is in the box above. The rest of this piece is about what the selection signals, and about what we have decided to do with the capital and credibility it provides.

What this signals about the AI funding stack

The standard venture model rewards traction, then revenue, then growth. It works for consumer software and for SaaS where product-market fit can be demonstrated in months. It works less well for AI infrastructure in regulated industries.

The sales cycle is long. A bank, a hospital system, or a government department does not buy AI infrastructure in a quarter. Security review, procurement compliance, internal model risk assessment, and regulatory approval add six to eighteen months before a paid deployment is signed. A founder optimising for revenue traction in year one cannot survive these timelines without runway that does not depend on revenue arriving on schedule.

Compliance is required before the first paying customer, not after. Cyber Essentials, ISO 27001, SOC 2, EU AI Act conformity assessments. These are gates, not optimisations. A regulated buyer will not pilot a system that lacks them. Building compliant infrastructure has to happen with capital that does not ask how many customers you have yet.

Defensibility comes from architecture, not network effects. There is no growth flywheel in audit infrastructure. The moat is in the cryptographic guarantees, the compliance mappings, the isolation properties, and the institutional trust accumulated over years of disciplined engineering. Capital that demands monthly growth metrics actively penalises the work that builds the moat.

Non-dilutive capital aligned with capability rather than traction is better matched to this category. Emergent Ventures sits inside a broader funding stack (UKRI in the UK, ARIA, NIST programmes in the US, sovereign tech funds across the EU) that is now meaningfully underwriting infrastructure work conventional VC cannot. We expect this stack to grow.

What we are building with it

The grant funds three commitments we are now making publicly.

Cyber Essentials Plus certification by Q3 2026. The basic Cyber Essentials self-assessment is in progress. The Plus tier requires independent technical audit. We are scoping with assessors now and expect certification to complete during Q3. This unblocks UK government procurement under CCS frameworks and is a baseline expectation for UK financial services pilots.

ISO 27001 audit pathway initiated in 2026, certification by Q2 2027. ISO 27001 is the standard a regulated enterprise buyer expects to see on a vendor security questionnaire. It cannot be rushed. The work begins this year with the gap analysis and ISMS scoping. The Stage 1 and Stage 2 audits follow on the standard timeline.

Open-source release of the EU AI Act compliance harness by end of 2026. The internal harness Hivemind AX uses to map system behaviour to Articles 9, 11, 12, 14, and 15, the same machinery that produced the compliance declaration we published in April, will be released under an open licence before year-end. This is the single most useful thing we can do for the broader category. Most teams building AI for regulated industries are reinventing this wheel privately. It does not need to be reinvented.

These are public commitments. We are writing them down so they can be checked.

Beyond the three, the grant funds continuation of the fine-tuning work begun on Cambridge Dawn under our UKRI AIRR allocation, conference presence (we are speaking at the AIRR User Day at Churchill College in May), and the operational runway to extend the design-partner programme into Q4 without compressing the engineering roadmap.

The asymmetry of regulated infrastructure

The infrastructure layer for AI in regulated industries is being built right now, in 2026 and the two years that follow. It will be built largely by teams who could afford to invest in compliance machinery, cryptographic guarantees, and audit infrastructure before they had a customer to bill.

This is the asymmetry the conventional venture model misprices. By the time a regulated AI infrastructure company has revenue traction sufficient to raise a Series A on standard terms, the technical work that determines whether the company can serve regulated customers at all has already been done. Or not done. A team that arrives at the regulated buyer’s procurement desk without Cyber Essentials, without an audit trail design, without a documented EU AI Act conformity position, does not get a second meeting. The work has to happen earlier than the venture timeline funds.

What the alternative funding stack does well is fund this earlier window. The selection is not based on which company has the largest pipeline. It is based on whether the technical and architectural work is serious enough to matter when the regulated infrastructure category consolidates.

We think the consolidation happens within twenty-four months. The teams building credibly now will be the teams the regulated buyers shortlist.

What is next

Hivemind AX is in open beta. We are running a small number of design-partner deployments with teams in fintech, healthcare, and government. Selection criteria for the design partner programme are published on the company page.

The fine-tuning work on Cambridge Dawn continues into Q3. A separate Insights piece on the methodology and findings will follow shortly, in particular what twenty thousand GPU hours of grid search across thirty-two LoRA configurations taught us about role-conditioned training in multi-agent orchestration.

For teams building AI systems where compliance, audit, and verifiable behaviour are not optional features, the contact form on this site reaches the founding team. If you would prefer a fifteen-minute conversation, the same page links a calendar.

Acknowledgments

Thanks to Tyler Cowen and the Emergent Ventures team at the Mercatus Center, George Mason University, for backing the work. Thanks to Elisabeth Reeve and the Cambridge Dawn team for the UKRI AIRR Rapid Access allocation that has powered the technical work this grant builds on. Thanks to the early reviewers and design partners who have been generous with their time during open beta.