Pramaana Labs has closed a $27 million seed round led by Khosla Ventures, with the capital earmarked for building a formal verification layer targeting AI deployments in legal and healthcare settings, according to TechCrunch and Crypto Briefing, both reporting on June 17, 2026.

The size of the round is notable for a seed stage. Twenty-seven million dollars at pre-product or early-product phase signals that Khosla — not a firm known for hedging its conviction — sees structural demand here, not a feature gap.

The problem Pramaana is attacking is well-defined, even if solutions have been elusive. Formal verification, borrowed from hardware and safety-critical software engineering, uses mathematical proof techniques to guarantee that a system behaves according to its specification under all possible inputs — not just the ones you tested. Applied to large language models and AI inference pipelines, that is a genuinely hard problem: the state space is enormous, model behavior is probabilistic rather than deterministic, and the "specification" of what a legal or medical AI should do is itself contested and jurisdictionally variable.

Healthcare and law are the two verticals where hallucination is least tolerable. A billing-code suggestion that is wrong costs money. A drug-interaction flag that is wrong, or silently absent, can cost a life. A contract clause that an AI misreads or fabricates exposes firms to liability that no indemnification clause in a vendor agreement will fully absorb. Practitioners in both sectors have been vocal about this — and regulators in the EU, the UK, and increasingly the US have begun treating AI outputs in high-stakes contexts as a compliance surface, not just a product feature.

The formal verification framing is worth unpacking. Traditional software verification — think TLA+, Coq, or Isabelle — operates on code whose logic can be symbolically enumerated. Neural networks do not reduce to that kind of symbolic enumeration cleanly. What "formal verification for AI" most plausibly means in practice is a hybrid approach: wrapping model inference with constraint-checking layers, using verified runtime monitors that assert output properties against a formal spec, or applying verified fine-tuning bounds so that a model's outputs remain within a provably safe envelope for a defined distribution of inputs. None of these fully solves the general problem, but each meaningfully narrows the tail-risk that makes enterprise procurement teams in regulated industries reluctant.

That reluctance is the market Pramaana is addressing. Large health systems and law firms have budgets for this. They also have compliance officers who can translate "formal verification" into procurement language — "auditable guarantees" — in a way that resonates with boards and insurers. The sales motion is not consumer-facing; it runs through the same channels as enterprise security and risk tooling.

Khosla Ventures has a pattern of backing infrastructure bets early when a wave of application-layer activity creates a new class of critical dependency. The logic here follows that pattern: as AI adoption in law and healthcare deepens, the verification and auditability layer becomes load-bearing. Getting in at seed gives Khosla positioning before that layer consolidates.

What Pramaana has not yet disclosed publicly — at least as of this reporting — is the specific technical architecture, any existing customer relationships, or the founding team's background. Those details matter enormously for assessing whether the formal-verification framing is substantive or marketing. Seed-stage opacity is normal, but the gap between "we are building formal verification for AI" and a shipping product that a hospital or law firm can put in front of their compliance team is wide. Worth watching closely as the company moves toward a product announcement.

The $27 million gives Pramaana runway to hire deep technical talent — formal methods researchers are not a large population — and to pursue the slow, relationship-driven enterprise sales cycles that regulated industries require. Whether the math holds up at production scale, across the model diversity that real-world deployments involve, is the question that will determine whether this seed round becomes a category-defining Series A story or a cautionary note about the gap between theoretical rigor and production reality.