Startup Pramaana Labs captures $27 million from Khosla Ventures for reliable AI with formal verification

Pramaana Labs, a startup that unites artificial intelligence to mathematical proof, has just raised $27 million in a seed round led by Khosla Ventures. The goal is to bring formal verification methods — traditionally used in hardware and critical systems — to the heart of AI models, especially in areas where errors can cost dearly, such as tax preparation, legal contracts and the development of new medicines.

Reliable AI: What is formal verification and why is it crucial?

Formal verification is a technique that uses mathematical logic to prove, exhaustively, that a system behaves exactly as expected. Unlike conventional software tests — which only cover a few scenarios —, formal verification ensures that there are no unexpected bugs or behaviors under any circumstances. For artificial intelligence, this means that a model can be mathematically attested as correct before being implanted in a real environment.

How does Pramaana Labs apply formal verification for reliable AI?

Startup develops tools that translate AI models — especially neural networks and learning systems by reinforcement — into a formal language that can be verified with theorem proofers. Instead of just testing with thousands of examples, Pramaana Labs allows developers to demonstrate, for example, that a tax pricing algorithm will never exceed a legal limit or that a drug screening system will never recommend a toxic combination. The company already operates in partnership with law firms and pharmaceutical laboratories in the United States.

Reliable AI: Why does formal verification matter to Brazil?

In Brazil, sectors such as tax law, compliance and health already begin to adopt assistive AI, but face resistance due to lack of guarantees of correction. A model that misses a tax calculation or recommends a dangerous drug can generate millionaire lawsuits. Formal verification, applied to models used by Brazilian companies, could drastically reduce these risks. Furthermore, the approach of Pramaana Labs echoes global regulatory initiatives — such as the European AI Act — which require greater transparency and security in high-risk systems.

Sectors benefiting most from AI with formal verification

Three priority areas for Pramaana Labs:

Right: verification of contracts and conformity analyses, ensuring that clauses are interpreted within predefined limits.

Drug discovery: validation of results generated by AI to avoid false positives or hazardous combinations.

Preparation of taxes: mathematical verification of automated tax returns and calculations.

Startup also plans to expand to finance and manufacturing, where reliability is equally critical.

Future of reliable AI: investment by Khosla Ventures in formal verification

Khosla Ventures, known for deep tech betting, saw at Pramaana Labs an opportunity to fill a gap in the AI market: the lack of formal guarantees. The $27 million check, in the Seed round, is one of the biggest visas ever for a company focused exclusively on formal verification. The expectation is that money is used to hire mathematicians and software engineers, as well as expanding the customer base.

Meanwhile, the market is still on the lookout. Pramaana Labs' approach can become a competitive differential in a scenario where AI-driven cyber attacks grow 4x in a year, and trust in intelligent systems is increasingly questioned.

Frequently Asked Questions about reliable AI and formal verification

What is formal AI verification?

It is a technique that uses mathematical logic to prove exhaustively that an AI model always behaves as specified, eliminating unexpected bugs and behaviors.

Who are the investors of Pramaana Labs?

The round was led by Khosla Ventures, one of the main venture capitalists of Silicon Valley, focused on high-impact and long-term technologies.

When will technology be commercially available?

Startup already operates in a pilot phase with customers in the USA and plans to launch a broad platform by the end of 2026, expanding to markets like Brazil in 2027.

---

Fonte original: TechCrunch