Don't Leave Your AI Intern Unsupervised

Across engineering, support, and infrastructure teams, autonomous agents are being deployed at speed. But, if LLMs are unable to differentiate between data and instructions, are we turning every log, ticket, message, and repository into a potential command channel?