Vercel CEO Guillermo Rauch has revealed an incident where an AI Agent, while using Opus 4.6 and OpenClaw, mistakenly generated a false GitHub repository ID despite knowing the correct project ID. This error led to the deployment of unrelated code on a user's server, as the random ID coincidentally matched a real open-source project. SlowMist CISO 23pds warned that as AI Agents become more widespread, new security challenges will emerge, including attacks on automated deployment processes through methods like AI search poisoning and search result manipulation.