Model Security Is the Wrong Frame – The Real Risk Is Workflow Security

According to The Hacker News, security focus is shifting from protecting AI models to safeguarding the workflows that surround them as AI copilots become embedded in daily work. The piece notes two Chrome extensions posing as AI helpers were caught stealing data from more than 900,000 users, and separate demonstrations showed prompt injections hidden in code repositories could trick IBM’s AI coding assistant into executing malware.

It emphasises that neither attack compromised the AI algorithms themselves, but exploited the context in which the AI operates, making the workflow the real target. The article argues that AI should be treated as a workflow engine, linking applications and automating tasks, which creates new integration pathways and enlarges the attack surface at every input, output, and integration point.

It suggests a shift in security strategy towards protecting the entire workflow with middleware guardrails and by treating AI agents like any other user or service, with restricted access and continuous anomaly monitoring. As a practical example, it highlights Reco as a leading platform that provides visibility into AI usage and enables guardrails at the workflow level. 15 January 2026.