# Harmonic expands its math-proving AI into C and Verilog code checks > Source: > Published: 2026-10-28 10:10:55+00:00 *Harmonic built Aristotle to prove Olympiad math problems in Lean. Now it wants the same model to prove your firmware has no undefined behavior before it ships.* Harmonic said on October 27 that Aristotle, the AI model it built to solve formal mathematics, can now verify software written in C and chip designs written in Verilog, according to a Reuters report on the announcement. It's a sharp turn for a company that spent its first two years chasing gold-medal performance at the International Mathematical Olympiad, and it's a bet that the same machinery that proves theorems can prove your code actually does what you think it does. That machinery is the point. Most AI code review tools work by pattern matching: they've seen a lot of buggy code before, so they guess where new code might also be buggy. Aristotle doesn't guess. It translates a specification into a mathematical claim, then tries to construct a formal proof, checked step by step in the Lean 4 proof assistant, that the code satisfies it on every reachable path. If no proof exists, you find out the property doesn't hold rather than hoping a language model's intuition catches it. For C, Harmonic says Aristotle can prove a program has no undefined behavior, the class of bug that compilers are free to handle however they like and that has produced some of the ugliest security incidents in computing history. For Verilog, the pitch is the same logic applied to hardware description language: prove a chip design matches its specification before it goes to a fab, rather than finding out after tape-out that a counter overflows under a corner case nobody tested. The reason this matters more for chips and firmware than for, say, a web app is simple: you can't patch silicon. A bug in a mobile app ships a fix next Tuesday. A bug baked into an ASIC means a redesign, a new mask set, and months of delay, often after the chip is already in a customer's hands. Firmware sits in the same uncomfortable spot, running below the operating system on devices that may never see another update. That's exactly the failure mode the industry has been living through this year, from UEFI firmware exploits to the kind of low-level flaws that turned up in Tailscale's SSH implementation. Catch it before fabrication, and it's a line in a proof. Catch it after, and it's a recall. [Claude Opus 5.5 helped verify what may be mathstory's biggest single day](https://startupfortune.com/claude-opus-55-helped-verify-what-may-be-mathstorys-biggest-single-day/) Anthropic says an internal Claude research model rederived the Exponential Time Hypothesis refutation on October 16, ten days before OpenAI's announcement, using roughly 14 million output tokens and no human steering. Reviewers David Gross and Arthur Jaffe said their involvement should not be taken as certification of the claims. - [Claude Opus 5.5 verified math breakthrough independently](https://startupfortune.com/claude-opus-55-helped-verify-what-may-be-mathstorys-biggest-single-day/) - [OpenAI Anthropic same day computer science proofs](https://startupfortune.com/claude-opus-55-helped-verify-what-may-be-mathstorys-biggest-single-day/) Harmonic isn't starting from zero credibility on the proving side. The company's own announcement, covered by Reuters, cites Aristotle resolving 96.8% of problems on VERINA, a code-verification benchmark built by researchers at UC Berkeley and Meta specifically to test whether AI systems can generate code alongside a provably correct specification. Aristotle proved 160 of 189 formal specifications correct and found 23 false, a result that outpaced the general-purpose language models tested when the benchmark was built. The company also pointed to a concrete customer story: the US fusion lab DIII-D, which Reuters reported had Aristotle rebuild part of its reactor-control software in four hours, starting from a single research paper and a 254-word prompt. That's a small, specific data point, not a sweeping claim, and it's the kind of detail that's actually checkable against Harmonic's own case studies rather than a marketing adjective doing the work alone. None of this happens without money behind it. Harmonic raised $120 million last November in a round that valued the company at $1.45 billion, backed by Nvidia's venture arm NVentures alongside Sequoia and Ribbit Capital, Reuters reported. Vlad Tenev, Robinhood's chief executive, co-founded Harmonic with Tudor Achim, and the Nvidia money is a tell in itself: a chip giant investing in a startup that formally verifies chip designs is not a coincidence, it's a supply chain bet. Aristotle for code is priced on usage, per Harmonic's announcement, rather than as a flat seat license, which suggests the company expects it to run as an automated check inside a build pipeline rather than a tool an engineer opens once a week. That's the real shift here. Formal verification has existed in chip design for decades, run by specialist engineers using tools like SystemVerilog Assertions, and Harmonic itself is hiring for exactly those roles. What's new isn't the math. It's compressing the proof-writing work that used to take a dedicated verification team weeks into something closer to a prompt and a few hours. Whether that holds up outside curated benchmarks and one fusion lab's reactor code is the open question. Formal verification is only as good as the specification you hand it, and writing a correct spec is its own hard problem that Aristotle doesn't claim to solve. Harmonic is betting chip and firmware teams will decide that risk is worth taking anyway, because the alternative, shipping silicon and finding out later, is worse. **Also read:** [SK Hynix says AI memory demand will outrun supply past 2030](https://startupfortune.com/sk-hynix-says-ai-memory-demand-will-outrun-supply-past-2030/) • [Nebius stock jumps 11% as tight GPU supply lifts the whole neocloud trade](https://startupfortune.com/nebius-stock-jumps-11-as-tight-gpu-supply-lifts-the-whole-neocloud-trade/) • [Claude Opus 5.5 Takes Top Spot as Artificial Analysis Launches Intelligence Index v5](https://startupfortune.com/claude-opus-55-takes-top-spot-as-artificial-analysis-launches-intelligence-index-v5/) *This article is posted in [AI News](https://startupfortune.com/category/ai/), check it out for more related stories.* ## Join the discussion [Open in the community →](https://startupfortune.com/community/) Almost there. Sign in and your reply posts straight away.