If you use generative AI tools, like ChatGPT and Google Gemini, you’ll know they produce confident-sounding results, often without any actual evidence to back them up. Midas is a new company, backed by investors connected to OpenAI, Tesla, and SpaceX, that is building mathematical verification tools that can check whether those results are actually correct.
The New York based company has closed a $10 million funding round led by Valor Equity Partners and Nova Global, alongside additional backers with experience building large scale technology companies.
Midas’ mathematical infrastructure verifies AI systems before outputs are allowed into real world use. Formal verification techniques are applied to AI models, training data, and reasoning processes, rather than evaluating results only after they are produced.
The Midas touch
The founding team consists of 11 medalists from the International Mathematical Olympiad and the International Olympiad in Informatics. These competitions admit only a small number of participants from each country each year and are widely regarded as among the most selective academic contests globally.
Team members have previously worked at Jane Street, Google, AWS, Nvidia, and Mercor. Academic backgrounds across the team include Stanford, MIT, Cambridge, Princeton, and Duke.
Shalim Monteagudo-Contreras, president and co founder of Midas, described the core limitation of current AI systems. “Modern AI produces fluent, convincing answers, but it cannot prove they are correct,” he said. “Midas is building the barrier between probabilistic outputs and real-world systems. We enforce correctness mathematically, so results are not inferred, argued, or hoped for, but proven before they are allowed through.”
Fluency and coherence are often mistaken for correctness, despite the fact that these qualities cannot be independently audited or verified.
Renzo Balcazar, chief executive and co founder, explained the issue in institutional terms. “Every human institution, from law to science to finance, runs on evidence,” he said. “Artificial intelligence is the first form of intelligence that operates without it.”
AI systems now generate outputs faster than humans can evaluate them, increasing risk as deployment expands into complex environments. Verification at the reasoning level becomes necessary once manual review is no longer realistic.
The platform introduces mathematical evidence into the core of AI workflows, verifying outputs, data pipelines, and internal reasoning steps. This approach applies in environments where mistakes are not an option.
John Stanton, vice president at Valor Equity Partners, said: “Verification is the final missing layer. This is not about probabilities, but proof. What sets Midas apart is its culture: a team trained to reject ‘almost correct’ answers and accept only what can be demonstrated.”
The funding is being used to translate formal verification research into production infrastructure. Initial deployment targets include biotech, defense, hardware design, financial systems, and underlying AI and cloud infrastructure.
You can find out more about Midas here.
What do you think about mathematical verification as a foundation for AI systems? Let us know in the comments.
