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

Reasoning models can be tricked into denial-of-service attacks with illogical prompts