{
  "id": 2195417,
  "title": "What Happens When the World is Run on Code No One Understands?",
  "url": "https://urgent.news/2026/08/20/what-happens-when-the-world-is-run-on-code-no-one-understands",
  "topic": "culture",
  "section": "Culture",
  "published": "2026-08-20T18:08:30.000Z",
  "source": {
    "name": "Time",
    "slug": "time",
    "url": "https://time.com/article/2026/08/20/what-happens-when-the-world-is-run-on-code-no-one-understands-/"
  },
  "original_language": "en",
  "account": "The International Congress of Mathematicians, held in Philadelphia for the first time in the U.S. since 1986, saw mathematician Jacob Tsimerman leave academia to focus on AI safety after receiving the Fields Medal, the highest honor in mathematics. Tsimerman, along with nine other mathematicians, was tasked with reviewing an AI model's claim to have refuted an 80-year-old mathematical conjecture. Their review translated the AI's argument into human-understandable mathematics.\n\nThe article highlights that the bottleneck that inhibits progress in mathematics and other fields is changing with the advent of AI. While discoveries are now seemingly endless, the scarcity lies in human verification. AI tools are expected to complement human ingenuity and expand knowledge and capabilities, but the infrastructure to review and certify discoveries was designed for human throughput. This constraint slows down innovation.\n\nAs AI advances, it is not only impacting mathematical discoveries but also the code that runs essential systems like hospitals, banks, and power grids. Several recent incidents, such as Anthropic's Mythos model uncovering unknown vulnerabilities in major operating systems and browsers, Microsoft's discovery of 90 critical flaws in a widely-used product, and Senator Mark Warner's claim that AI broke into nearly all of the U.S. classified systems in hours, demonstrate the growing threat posed by AI in these domains.\n\nThe issue extends beyond mathematics, as AI is also writing much of the code that powers our critical infrastructure. Developers can now \"vibe code\" using generative AI tools, which allows for rapid development but also leads to vast amounts of code that many do not fully comprehend. These AI systems that accelerate discovery can also help manage the resulting information overload, provided they are combined with the precision and rigor of mathematics. Formal methods, which involve proving that software mathematically performs as intended, can help verify AI output and improve software security.\n\nHowever, formal verification is not a panacea. It cannot prevent phishing or ensure the safety of complex human systems. Its main role is to eliminate certain classes of errors in critical software. For systems where failure can lead to harm to millions, mathematical rigor should be a crucial step in adopting AI-generated code.\n\nThe Leiden Declaration, endorsed by the International Mathematical Union, emphasizes that AI threatens the verifiability of proofs. Instead of viewing mathematics as a threat, the article advocates for viewing it as the infrastructure to make generative AI more trustworthy. To achieve this, the U.S. should treat mathematical rigor as a national mission, investing in infrastructure for verified software, standards, benchmarks, better tools for checking updates, and educational programs that bridge mathematics, computer science, engineering, and national security.",
  "summary": "Mathematical verification must be a national mission in the AI era, write Patrick Shafto, Ken Ono, and Scott Duke Kominers.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}