Juniper Ventures backs the checks before the wrecks.

Consider the insurance policy. It is an unexciting document with an extraordinary job: turning uncertainty into something a business can live with. Artificial intelligence has produced plenty of demonstrations of what machines can do. Juniper Ventures is interested in the machinery that makes those demonstrations usable when the consequences become real.

The San Francisco firm launched its first fund in 2024. Its argument is straightforward: as AI acquires more responsibility, someone must test it, constrain it, establish who is using it and decide who carries the risk. Those jobs create businesses. Juniper backs the founders building them.

The interesting part is everything around the model

Take three names from its portfolio. Goodfire works on interpretability, helping people understand and modify the internal workings of neural networks. Apollo Research studies misaligned behavior, including systems that appear cooperative while pursuing other goals. Artificial Intelligence Underwriting Company, or AIUC, combines standards, audits and insurance. Each approaches trust through a different door.

This is what Juniper means by assurance infrastructure. Its interests extend from energy, hardware and networking through cloud security, model evaluation and runtime monitoring to authentication, compliance and insurance. The firm is looking across the chain of dependencies that connects an intelligent system to a customer willing to use it.

The distinction matters to a founder. A company selling proof that an AI workload ran in a protected environment has different customers and technical requirements from a company selling content moderation. Both can fit Juniper’s thesis. The common question is whether the product removes a concrete obstacle to using AI safely.

A standards business writes a check

In May 2026, DNV Ventures announced a $2.5 million commitment to Juniper Ventures Fund II. DNV and Juniper had spent about six months collaborating on AI safety and industrial AI governance. DNV described the investment as an extension of its assurance-of-AI mandate.

There is a useful clue here. DNV brings experience with standards in industries where failures have physical consequences. Juniper brings exposure to young assurance companies. The partnership puts an established assurance organization alongside a venture investor backing the tools that might make AI easier to trust.

“AI assurance is no longer optional”Kaare Helle · DNV Ventures · May 2026

That commitment went into an investment fund. It is not Juniper’s revenue, a valuation or the total size of Fund II. At the announcement, DNV identified Apollo Research, Alien Labs and Workshop Labs as the fund’s first three investments. The examples span model behavior, identity and secure AI environments.

The founders behind the founder help

Nick Fitz and Griff Bohm are Juniper co-founders. Fitz previously co-founded and led Momentum, an AI donor-engagement company acquired by Virtuous in 2025. Bohm describes having founded three businesses and sold two. Their pitch to entrepreneurs includes help with fundraising, hiring, growth, product and governance, along with culture and well-being.

These are recognizable early-company problems. A research breakthrough does not recruit a first salesperson. A financing round does not settle a difficult governance decision. Juniper’s stated support reaches into the operating work that remains after the wire arrives. Its investment proposition depends partly on being useful there.

The current website states a check range of $1 million to $100 million. Its portfolio includes pre-seed investments and later participation, including Goodfire’s Series B. For founders, that means the firm’s public ambition extends beyond the first institutional check, even as early company-building remains central to its identity.

The customer question comes first

Bohm’s May 2026 essay about starting a new AI lab offers a sharper view of the investment filter. He reports seeing nearly 500 lab decks. His requirements include a novel technical approach, credible experience, a CEO capable of raising capital and a specific customer. The last requirement gives the others commercial purpose.

A founder can borrow this discipline immediately: name the capability, identify the application it unlocks and describe the customer evidence. A claim to build a better model leaves the buyer doing too much imaginative work. A defined use case gives both founder and investor something they can test.

There are two audiences in this business. Founders need money and an investor who understands the work. Limited partners supply capital to the funds that back those founders. Juniper competes for promising companies with specialist safety investors and broader AI funds. Its distinction is the assurance focus, paired with practical operating help. A founder should judge that combination against the help the company actually needs, whether that is recruiting researchers, finding buyers or financing a long technical development cycle.

The conditions matter. Technical differentiation must survive competition. A customer must care enough to pay. A research-heavy company needs financing long enough to reach its next result. Juniper’s thesis makes assurance a place to look; it does not make every safety project a venture business.

Some useful things will never have a buyer

Juniper’s affiliated Foundation addresses that last problem explicitly. Its fellowship advertises $100,000 grants, with no equity taken, for proven builders working full time on AI safety, biosecurity or cybersecurity. The distinction is practical: shared testing infrastructure and other public goods may matter even when no customer can support them.

The commercial fund also publishes the AI Assurance Technology Report, maintains a tracker of emerging AI labs and runs a safety-and-assurance jobs board. Its governance writing asks founders to consider how a mission survives new investors and future ownership. These resources make parts of Juniper’s thinking available beyond its portfolio.

The useful takeaway is a buying question. Before giving an AI system more responsibility, ask what evidence would make that decision defensible. A test, a hardware guarantee, an audit or an insurance policy might supply part of the answer. Juniper is betting that founders can turn those answers into companies.