Claude Flaw Automatically Sends Malicious Prompts to AI Agents

When combined with another exploit, the "PromptFiction" vulnerability, which has been fixed, could have enabled an end-to-end attack on a targeted system.

Published: 2026-07-15