Theorem wants to stop AI-written bugs before they ship — and just raised $6M to do it

Unlocking the Future of Software Development with AI

As artificial intelligence revolutionizes software development, a pioneering startup is placing a bold bet on the future. The next challenge in the industry won’t be about writing code; it will be about trusting it.

Theorem, a San Francisco-based company born out of Y Combinator’s Spring 2025 cohort, recently announced securing $6 million in seed funding. Their mission? To develop automated tools that ensure the accuracy of AI-generated software. Leading the investment round is Khosla Ventures, along with support from Y Combinator, e14, SAIF, Halcyon, and notable angel investors like Blake Borgesson and Arthur Breitman.

The urgency of this endeavor is evident as AI-powered coding assistants, such as those from GitHub, Amazon, and Google, are generating vast amounts of code annually. The rapid adoption of AI in enterprises has raised concerns about verifying the functionality of AI-generated software effectively, highlighting a critical “oversight gap” that jeopardizes essential infrastructure like financial systems and power grids.

“We’re at a tipping point,” remarked Jason Gross, co-founder of Theorem, emphasizing the overwhelming volume of AI-generated code surpassing human review capabilities. “Reviewing 60,000 lines of code manually is beyond our capacity.”

Theorem’s Innovation in AI Code Verification

Theorem’s groundbreaking technology merges formal verification—a method for proving software correctness—with AI models trained to automatically generate and validate proofs. This revolutionary approach streamlines a process that traditionally demanded years of specialized expertise into a timeframe of weeks or even days.

Formal verification, although an established practice for critical applications like avionics systems and cryptographic protocols, was previously cost-prohibitive for mainstream software development due to its labor-intensive nature. Gross, drawing from his experience developing verified cryptography code at MIT, underscores the transformation enabled by AI, making software verification economically viable.

By employing “fractional proof decomposition,” Theorem’s system optimizes verification resources based on the significance of each code component, rather than exhaustively testing all possible scenarios. This strategy recently detected a bug overlooked during testing at Anthropic, demonstrating the effectiveness of Theorem’s approach in identifying critical issues efficiently.

Real-World Impact of Theorem’s Technology

Collaborating with clients in AI research, electronic design automation, and GPU-accelerated computing, Theorem has showcased the practical value of its technology. In one instance, the startup transformed a 1,500-page specification into 16,000 lines of reliable code, resolving performance challenges without manual intervention. This success story underscores the power of compact executable specifications paired with equivalence-checking mechanisms in ensuring flawless software implementations.

As AI increasingly permeates critical infrastructure, concerns about the security of AI-generated software intensify. Gross emphasizes the need for “asymmetric defense” in software security to counter the escalating threats posed by AI-enabled vulnerabilities. Theorem’s focus on verification as a cornerstone of software security reflects a proactive stance in safeguarding digital ecosystems against evolving risks.

Looking Ahead: The Future of AI-Powered Software Verification

Theorem intends to leverage its recent funding to expand its team, enhance computational resources for training verification models, and venture into diverse sectors like robotics, renewable energy, cryptocurrency, and pharmaceuticals. The startup’s emergence signals a paradigm shift in evaluating AI coding tools, emphasizing the imperative of ensuring software accuracy alongside speed.

Theorem’s vision extends beyond mere productivity gains to a realm where mathematical proof underpins the safety and reliability of AI-generated code. As AI continues its exponential advancement, Theorem stands at the forefront of enabling oversight and control in a rapidly evolving digital landscape.

Amidst the era where machines craft the code, Theorem emerges as a beacon of assurance, ensuring the integrity of AI-driven software that shapes our interconnected world.

Leave a Reply

Your email address will not be published. Required fields are marked *