The pace of discovery driven by artificial intelligence is outpacing humanity's ability to verify its findings, creating a potential bottleneck for innovation and a national-scale engineering challenge, experts warn.
Mathematicians and computer scientists are calling for mathematical verification to become a national mission in the era of AI. They argue that while AI is accelerating discovery, the existing infrastructure for vetting and certifying these breakthroughs was designed for human limitations and is now the primary constraint on progress. The solution, they propose, lies in formalization—translating AI outputs into precise forms that can be automatically checked for correctness.
This challenge extends beyond theoretical mathematics. The code that underpins critical infrastructure, including hospitals, banks, and power grids, is increasingly complex and generated or analyzed by AI. In April, Anthropic reported that its Mythos model identified previously unknown vulnerabilities in major operating systems and browsers, prompting alerts to fifty organizations to address the issues. Shortly after, Microsoft engineers discovered 90 critical flaws in a widely used product. Citing the NSA director, Senator Mark Warner noted in June that a similar tool had penetrated nearly all classified U.S. systems within hours, not weeks.
These incidents highlight a growing concern: as the world becomes more reliant on code that even its creators may not fully understand, the potential for undetected vulnerabilities increases. The authors advocate for building the infrastructure to make verification routine, viewing it as a crucial step to ensure the safety and reliability of systems that power modern society.