Claude extension flaw enables AI action hijacking

A security flaw in a Claude extension allows attackers to inject malicious prompts into AI agents, hijacking their actions. The vulnerability enables unauthorized command execution through crafted inputs, posing risks to automated workflows and data integrity.