Palo Alto Networks’ Nikesh Arora is building a defense against the dark side of AI

In April, Nikesh Arora applied a simple but unsettling test to his $300 billion cybersecurity company, Palo Alto Networks: He turned loose on the company’s internal infrastructure an unreleased version of a terrifyingly powerful new AI model called Mythos, created by the artificial intelligence company Anthropic. Arora’s team prompted it to find vulnerabilities in the company’s systems. Claude Mythos 5, Anthropic’s frontier AI model for advanced coding and cybersecurity work, was so effective at hacking into systems that the U.S. government has sought to limit its release. It even temporarily imposed export controls on Mythos and a related model called Fable in June, a move that forced Anthropic to disable the model for all users. The government subsequently allowed the company to restore access to Mythos for a small number of vetted U.S. users, and later allowed Anthropic to make it available to select users in 15 other countries. At Palo Alto Networks, the results of the controlled test were troubling. Mythos surfaced weaknesses in Palo Alto’s systems at a speed and scale that Arora immediately knew would change cybersecurity permanently. He also realized that Mythos—and competing models from Anthropic rival OpenAI—had created “the best marketing moment for the cybersecurity…