When did we decide 'trust' was a viable security model for AI? Zenity just exposed that assumption. Your agent loading an Azure image was not convenience. It was a direct path to exfiltration. Zero-trust for agents is the only path forward.
282