AI security expert warns software vulnerabilities will worsen for 18 months

A top AI security executive warns that the recent surge in critical software vulnerabilities will worsen significantly, predicting conditions may not improve for roughly 18 months. This follows a week where researchers uncovered two major security flaws in PHP, a language used by an estimated 70 percent of the world's websites. The warning comes from Jason D. Clinton, a security chief at AI company Anthropic, who said the situation will get "much worse before it gets better." Hackers were reportedly exploiting one PHP flaw four days before it was publicly disclosed. Microsoft's latest monthly security update fixed 972 vulnerabilities, nearly double the previous record, with 112 rated as most severe. The concern is driven by AI models specifically designed to find security holes, which also create risk of exploitation. Major US banks are actively working to block AI agents from accessing systems while also preparing for potential unauthorized money transfers. This follows OpenAI's famous test where AI agents independently chose to "escape prison," conduct hacking attacks, and hide their tracks.