Amazon's New Tool Makes AI Agents Ask Your Permission First

Amazon Web Services introduced a new feature called the Consent portal for its AI-building toolkit, Bedrock AgentCore. Think of it as a permission slip page for AI agents — the software that can actually do things for you, like read your code or send messages. When an AI assistant wants to touch your GitHub account, you'll get a simple web page: sign in with your company login, see exactly what the AI is asking for, and check only the boxes you're comfortable with.

Until now, any company wanting this setup had to build the whole thing from scratch — the sign-in page, the redirects, the secure storage, and the tracking of who approved what. That's the kind of plumbing IT departments dread. Amazon now handles all of it. The tricky part it solves is called "session binding," which just means making sure each approval stays attached to the right person, so your colleague's permissions never bleed into yours.

This is aimed at developers using AI coding assistants like Cursor, Claude Code, and VS Code, where the assistant wants to file bug reports or check repositories for you. Each person connects once and then forgets about it — no repeated pop-ups asking to log in. Administrators set up the portal, share one link, and everything is trackable in Amazon's activity logs, which is good news for anyone working in a regulated industry like finance or healthcare.

The catch: this only works if your company already runs on Amazon's cloud and has a corporate login system in place. It's not something you can switch on at home, and it won't help small teams on Google or Microsoft. For everyone else, it's a sign of where things are heading — your AI will soon have to ask nicely before touching anything.