Claude flaw automatically sends malicious prompts to AI agents

The 'PromptFiction' vulnerability in Claude could automatically inject malicious prompts into AI agents, potentially enabling end-to-end attacks. The flaw has been fixed by Anthropic.