A one-pixel hidden text block on a webpage successfully tricked AWS Kiro, an AI coding IDE, into rewriting its own configuration and executing arbitrary code without proper approval. This vulnerability bypassed the human-in-the-loop security model by injecting malicious instructions that the AI misinterpreted as legitimate setup tasks. Although AWS has patched the issue in version 0.11.130 and added protected file paths, the incident reveals significant risks when AI agents can modify their own execution environments. Businesses must now treat such AI tools as high-risk assets that require strict platform controls, regular updates, and isolation from untrusted web content.