From 247WallSt
“The Nuclear Weapons of Cybersecurity”: Why Treasury Just Warned Banks About AI’s New Power
By Omor Ibne Ehsan
It is unusual for the Treasury Secretary to call the heads of the largest Wall Street banks into a room to discuss a software system. It is even more unusual when the Federal Reserve Chair joins him. Treasury Secretary Scott Bessent and Fed Chair Jerome Powell recently gathered the CEOs of major Wall Street banks to warn them about Anthropic’s newest AI platform, Mythos, an artificial intelligence system reportedly so capable at hunting down software vulnerabilities that the company has only handed a preview to a handful of big tech and finance firms so they can patch holes before the rest of the world catches up.
That is the backdrop for a striking line from Steven Weber, the retired UC Berkeley professor who led the Center for Long-Term Cybersecurity. “Zero-day exploits are the nuclear weapons of the cybersecurity world,” Weber said. He argues that the rapid coding ability of frontier large language models has made this moment inevitable.
A zero-day is a software flaw the vendor has not yet discovered, which means defenders have zero days to fix it before an attacker can use it. Historically, finding one took elite human researchers weeks or months. AI systems trained on vast code corpora can now sift through software at machine speed, and that is the capability that has regulators alarmed.
OpenAI’s latest model, GPT-5.4 cyber, is raising similar concerns, and both Mythos and GPT-5.4 cyber excel at detecting zero-day exploits. The same skill that lets a defender harden a trading system lets an attacker compromise it. That is the dual-use problem in one sentence...
Steven Weber is a professor emeritus of the I School, retiring in 2021. He previously served as the faculty director at the Center for Long Term Cybersecurity (CLTC).
