Haokun Qin Proposes Formal Verification for GenAI Compliance

Haokun Qin, cofounder of Gale, proposes the adoption of formal verification to ensure Generative AI compliance and safety for enterprises. Qin argues that using mathematical proofs to verify AI workflows can prevent revenue erosion and costly regulatory fines, citing legal risks such as the EU AI Act and a $365,000 settlement involving the Equal Employment Opportunity Commission over AI bias.

The proposed framework utilizes finite-state models created with tools like TLA+ or PROMELA, encoding safety requirements in temporal logic and employing SMT solvers for exhaustive state analysis. To reduce implementation barriers, Qin suggests a phased roadmap that begins with pilot projects and progresses toward independent audits and CI/CD integration.

Qin also highlights the role of emerging technologies in making these processes more accessible and computationally efficient. He points to zero-knowledge virtual machines from RISC Zero and Jolteon, as well as Aleo's zero-knowledge DSL, as tools that can integrate compliance checks directly into AI workflows.