AI Models Overthink Problems—and It’s a Security Risk

Source: IEEE Spectrum AI·2026-07-08 19:00Publish time unverified

Large language models (LLMs) that can think through problems step-by-step have significantly increased the scope of tasks that AI can tackle. But new research suggests these reasoning capabilities also introduce a critical vulnerability that could allow attackers to slow these systems to a crawl.