Computer scientist Tudor Achim and his company, Harmonic, are developing a mathematically provable AI system designed to eliminate hallucinations [1].
This approach seeks to solve the reliability crisis in artificial intelligence. By grounding AI in formal verification, the company intends to move beyond probabilistic guesses toward a system where errors are logically impossible.
Achim presented this vision for "mathematical superintelligence" during a TEDAI conference in San Francisco on Oct. 21, 2026 [1]. He said that the goal is to realize the dream of Gottfried Wilhelm Leibniz, who envisioned a logical framework where errors could be avoided entirely [1].
To fund this research, Harmonic announced a Series C financing round of $120 million [2]. The investment round, which included Ribbit Capital, brings the company's post-money valuation to $1.45 billion [2].
Traditional large language models often struggle with factual accuracy because they predict the next likely word rather than reasoning through a formal proof. Harmonic's proposal suggests that AI should instead be built on a foundation of formal mathematics to ensure every output is verified [1].
"Our goal is to build the first mathematically provable AI system that eliminates hallucinations," Achim said [2].
Ribbit Capital emphasized the potential for this technology to transform technical fields. A partner at the firm said that mathematical superintelligence represents the next frontier for reliable scientific discovery [2].
While current AI tools are widely used for creative tasks, their tendency to invent facts has limited their use in high-stakes environments, such as medicine or engineering, where a single error can be catastrophic [1].
“"Our goal is to build the first mathematically provable AI system that eliminates hallucinations."”
The shift toward formal verification marks a departure from the 'black box' nature of current generative AI. If Harmonic successfully integrates Leibniz's logical frameworks with modern computing, AI could transition from a tool of probability to a tool of proof, making it viable for critical infrastructure and scientific research where accuracy is non-negotiable.
