Amazon's AI Tools Now Check Your ID Before Every Task

Amazon just published a detailed guide for companies that plug AI assistants into their private data. The setup involves MCP (a standard plug that lets AI reach tools, databases, and apps), and Amazon Quick, Amazon's AI assistant. The problem: when an employee signs in, the company knows who they are but not what they're allowed to do. If that one login is too broad, or someone steals it, the AI could pull records the person should never see.

The fix is what Amazon calls "defense in depth" — several checkpoints in a row rather than one. Every time the AI tries to do something, it passes through four gates: did the person use two-factor authentication, are they logging in from an approved country, does their job role allow this kind of action, and is this specific tool permitted? A reader, for example, can look up risks in a company register but cannot delete them; an administrator can do both. Every change gets logged permanently for auditors.

Why should you care if you don't work in tech? Because this is the plumbing that decides whether your employer's AI can be trusted with your HR file, medical claim, or bank record. Companies in finance, healthcare, and government are already under pressure to prove they control who sees what. Without this, many of them simply won't turn AI on — which slows down the useful stuff, like asking an assistant a question about a policy instead of emailing three people.

The catch is that this is do-it-yourself. Amazon is not shipping a magic switch; it's a technical walkthrough assuming the company has already built the underlying pieces and has an identity provider like Microsoft Entra ID. A single misconfigured permission can still undo the protection. Expect this to become a standard checklist — and a selling point — over the next year.