Harmonic, the artificial intelligence startup co-founded by Robinhood chief executive Vlad Tenev, said on Tuesday, October 27, 2026, that it is expanding its Aristotle model from mathematical proofs into software and chip design verification, bringing its proof-based approach to two of the places where a single error can be most expensive. The company said Aristotle can now verify software written in C and chip designs written in Verilog, a move that takes a system built to check mathematical reasoning and points it at the code and hardware that run the modern world. Harmonic is developing what it calls mathematical superintelligence, a reasoning-focused form of AI aimed at reducing hallucinations and factual errors.
From Proofs To Programs
Aristotle began as a mathematics machine. Rather than answering with the most plausible-sounding text, it works toward results that can be checked step by step, in the way a proof can be checked line by line. That discipline is the heart of Harmonic’s pitch: where a conventional chatbot may confidently produce a wrong answer, a proof-based system is supposed to show its working.
The expansion applies that discipline to engineering. Software verification asks whether a program actually does what it is supposed to do, rather than merely whether it runs without crashing in a demo. Chip design verification asks the same question of hardware, before a design is manufactured, when fixing a flaw becomes far slower and more expensive.
C and Verilog are deliberate first targets. C remains the language of operating systems, embedded devices and much infrastructure software. Verilog is a hardware description language used to design digital circuits, so a verified Verilog design is a claim about a chip’s behaviour, not just about software. By covering both, Harmonic is aiming at the full stack, from the code that runs on a chip to the chip itself.
The PHP Warning Shot
The announcement lands as AI’s role in finding software flaws draws intense scrutiny. Reuters notes that researchers recently used the technology to uncover critical bugs in PHP, prompting a global rush to patch the vulnerabilities. The episode showed both sides of the new reality: AI-assisted analysis can surface flaws buried in widely used code, and once those flaws are public, the race to patch them begins immediately.
CISA has since added the PHP flaws to its known exploited vulnerabilities catalog, with a patch deadline for federal agencies, a reminder that discovery is only the first step. The harder problem is proving, before release, that the next critical flaw is not already sitting in a codebase. That is the gap verification tools are meant to close.
A Fusion Lab Test Case
Harmonic also pointed to a striking early example from scientific computing. The United States fusion lab DIII-D said Aristotle rebuilt part of its reactor-control software in four hours, from one research paper and a 254-word prompt. Reactor control is exactly the kind of high-stakes software where correctness matters more than speed of writing.
The example is a single case, not proof that the approach works everywhere. But it illustrates the intended use: a domain expert supplies the goal and the source material, and the system produces code that can be checked rather than merely trusted. For laboratories, aerospace teams and chip designers, that distinction is the whole point.
The Backing Behind The Bet
Harmonic is not a small research project. Backed by Nvidia’s venture capital arm, Sequoia and Ribbit Capital, the company raised $120 million in a funding round last November that valued it at $1.45 billion. That backing puts it in a growing group of AI companies betting that the next advantage will come not from a bigger general-purpose chatbot, but from systems that can be trusted with tasks where errors are costly.
The bet also fits a wider shift in how AI is being judged. Independent safety tests of AI models are drawing closer attention, and the question of whether an AI system can be kept under human control has moved from theory to procurement checklists. A system that proves its answers, rather than asserting them, speaks directly to buyers who need audit trails, not just impressive demos.
Why It Matters
Most AI headlines this year have been about bigger models, faster chips and larger funding rounds. Harmonic’s expansion matters for a different reason: it targets trust. If AI is going to write production software, design hardware and control physical systems, the industry needs ways to check its work that do not depend on the same system’s say-so. Formal verification is one of the few methods that offers that kind of independent check.
The open question is scope. Verifying a piece of C code or a Verilog design is not the same as verifying an entire system, with its drivers, networks and human operators. Customers will judge Harmonic on how much real-world code and hardware it can handle and how often its proofs survive contact with messy, legacy systems. If it clears those bars, proof-based AI could become a standard part of how critical software and chips are built.
Background
Harmonic was founded to pursue mathematical superintelligence, an AI that reasons with the rigour of mathematics rather than the approximations of language prediction. Its Aristotle model first drew attention for mathematical problem-solving, and this expansion extends that approach from theorems to engineered systems that depend on AI-generated code.
Source: Reuters report on Harmonic
0 Comments