The problem with insuring robots that fly is that nobody quite knows how to price the risk.
Enter Valgo, a San Mateo outfit so young it barely shows up in startup databases, yet already threading a needle between two of the thorniest challenges in autonomous systems: proving they're safe, and convincing someone to underwrite them. Founded by Stanford PhDs whose academic pedigrees run through collision-avoidance systems and Bayesian validation theory, Valgo started life building algorithmic safety tools for autonomy developers. Then came the pivot.
It happened quietly, in the months before European aviation regulators published concrete frameworks for AI trustworthiness in flight systems. Suddenly, the same simulation-based validation methods that could help engineers hunt down rare failures in autonomous aircraft had another buyer: insurance companies desperate for actuarial footing in a market they don't yet understand.
"Risk quantification platform to insure physical AI," the company's Y Combinator profile now reads—a phrase that would have sounded like science fiction five years ago.
When Certification Meets Cold Economics
Valgo's shift speaks to something larger than one startup's strategy. As machine learning seeps into everything from delivery drones to urban air taxis, the question of liability is moving from theoretical to urgent. Insurers can't price what they can't model. And they can't model what autonomy developers can't rigorously validate.
The company didn't so much abandon its original product as discover who actually needed to pay for it. Early descriptions—still visible in search caches and startup directories from months back—emphasized "algorithmic safety validation tools" aimed at accelerating certification. Black-box methods, agnostic to underlying models or simulators, designed to efficiently surface rare, realistic failure scenarios across autonomous vehicles, aviation, robotics, even space systems.
That core technology remains. But instead of selling directly to developers navigating the gauntlet of regulatory approval, Valgo now positions itself as the translator between their simulations and an underwriter's spreadsheet. Companies building autonomous aircraft run their own sims; Valgo ingests aggregate statistics via API and spits out loss estimates that insurers can—theoretically—use to set premiums.
Whether insurers will trust those estimates is another matter entirely.
The Stanford Validation Curriculum, Productized
Both founders bring deep technical credentials, the kind forged in academic labs where safety isn't marketing copy but mathematical proof. Robert J. Moss has extensive background in collision-avoidance systems and autonomous safety validation. Sydney M. Katz holds a Stanford PhD in aeronautics and astronautics, focused on safe machine learning, and currently co-teaches a course at Stanford on validation of safety-critical systems.
The winter syllabus covers optimization-based falsification, importance sampling, reachability analysis, model checking. Applications run from air traffic control to full autonomy. Moss and Katz are co-authoring Algorithms for Validation, a book project that essentially formalizes years of research into methods the startup is now trying to commercialize.
Valgo's third team member rounds out the picture: a Stanford Graduate School of Business Sloan Fellow with insurance leadership experience and a reported track record in M&A. On paper, it's the kind of team that can speak both technical and actuarial dialects—assuming those two worlds are ready to speak the same language.
The company operates as a Public Benefit Corporation, a legal structure that allows mission statements about improving critical system safety "for users, designers, operators, regulators, and their communities" to coexist with revenue targets. The insurance economics angle doesn't dilute that mission, the founders would argue. It just monetizes a validation layer autonomy developers need regardless of who's writing the checks.
Regulatory Winds, Maybe Tailwinds
European aviation authorities have been moving, however tentatively, toward structured frameworks for demonstrating AI system trustworthiness. Recent regulatory proposals signal an emerging consensus: Level 1 and Level 2 machine learning systems in aviation—broadly, assistance and human-AI teaming—will need more than good intentions and beta testing. They'll need evidence.
That's the regulatory environment Valgo is betting on. Across the Atlantic, the FAA has been updating its own AI safety materials, emphasizing learned-AI assurance through consensus standards, guidance documents, and staged deployments starting with lower-criticality applications. None of this constitutes a full roadmap yet, but it does suggest that the "move fast and break things" ethos has limited runway when breaking things happens at altitude.
For insurers, regulatory clarity—even the promise of it—makes underwriting criteria less speculative. Loss estimates derived from rigorous simulation methods become more defensible when regulators have sketched, in even rough form, what "trustworthy" AI actually requires. Valgo wants to be the platform supplying those estimates at scale, once the market matures enough to pay for them.
Whether it will remains an open question.
A Sparse Competitive Landscape
Valgo enters a field with adjacent players but few direct analogues. Daedalean, a Zurich-based company developing certifiable AI avionics, has collaborated extensively with European regulators and co-authored neural network assurance frameworks. Applied Intuition offers validation toolsets for automotive ADAS and autonomous driving systems, and has expanded into defense and broader "physical AI" applications. Legacy players like Ansys and MathWorks provide certification toolchains that support existing compliance standards, but don't focus on rare-event discovery or insurance-grade risk quantification—the specific niche Valgo is carving.
The founders haven't disclosed funding details beyond their Y Combinator batch participation. Demo Day is scheduled for late March, though no splashy seed round announcements have emerged yet. The company's GitHub organization lists a contact email but no public repositories yet.
What is public: the technical lineage. Collision-avoidance research. Xwing's autonomy division, acquired by Joby Aviation on June 4, 2024. Stanford's validation coursework, now being taught to a new generation of engineers who may one day be Valgo's customers.
Algorithms as Actuarial Tables
The deeper premise here is that autonomy's bottleneck isn't just technological—it's financial. Even if engineers can build systems reliable enough for certification, those systems won't scale commercially unless someone will insure them at rates that don't strangle the business model. And insurers, famously conservative about risks they can't quantify, won't underwrite at reasonable rates without better tools.
Valgo is betting that the answer lies in algorithms: Bayesian methods for black-box safety validation, importance sampling to surface edge cases, statistical frameworks that translate simulation runs into loss distributions. Whether those algorithms can convince a risk committee at a major insurer to sign off on a seven-figure policy—that's the real validation test.
As aviation autonomy inches from prototype to commercial deployment, the question isn't just whether AI systems can be made safe enough. It's whether they can be insured profitably, at scale, without waiting decades for actuarial tables to fill in naturally. Valgo is placing a quiet, technically sophisticated bet that better math can collapse that timeline.
Time will tell if the underwriters agree.
