AWS Now Lets AI Assistants Answer Questions from Data in Other Accounts

Companies often keep their data in one secure place and their AI assistants in another. That separation is great for security, but it makes it hard for the AI to actually use the data when answering employees' questions. AWS just showed how to bridge that gap safely, without copying any sensitive information from one storage account to another.

The key idea is simple: the AI doesn't bring the data to itself. Instead, it asks for a temporary, narrowly scoped permission badge (called an IAM role) from the account that owns the data. With that badge, it can call a special AI service that reads the underlying data, turns the user's question into a database query, and returns a natural-language answer — complete with citations. After the answer is given, the badge expires.

AWS describes two ways to build this. One uses a code-based coordinator that runs the AI step by step. The other uses a more managed approach where AWS handles part of the routing through a gateway. Both produce the same result: a secure question-and-answer experience for your employees, using data that never leaves its original secure account.

The catch: not every use case needs this level of complexity. AWS advises that if you just want the AI to retrieve raw documents, you can use a simpler method. And if you only need one fixed answer without multi-step reasoning, you might not need a full agent at all. The extra machinery is only worth it when users ask open-ended questions that require the AI to reason across different data sources.