Saturday, November 21, 2026
Translate
PrivacySubmit newsShare your story Newsletter Advertise
Breaking
Cyber news

AWS API Lets Developers Scope Down Permissions for AI Agents

Amazon Web Services launched GetScopedCredentials to let workloads give tasks and AI agents narrower access than their original sessions.

AWS API Lets Developers Scope Down Permissions for AI Agents
Image: Amazon Web Services

This gives developers a much tighter leash for autonomous workloads. Amazon Web Services released GetScopedCredentials, a new AWS Security Token Service API designed to let existing temporary security credentials reduce their permissions, shorten session durations, or both.

How GetScopedCredentials Works

According to Amazon Web Services, developers can call GetScopedCredentials using temporary security credentials from aws login, an Amazon EC2 instance profile, or an AWS Lambda function. Each call accepts an inline scope down policy, a shorter session duration, or both, ensuring the resulting permissions are the intersection of the session effective permissions and the policy.

The feature is available in all AWS commercial Regions, AWS GovCloud (US) Regions, and the China Regions with a dedicated request quota of 200 requests per second per account in each Region at launch. AWS states that a maximum of three scope down policies can apply to a session.

The Cyber Brief. Every weekday morning: what is being exploited, what to patch and who was hit, in one short email. Free.

Use Cases for AI Agents and Multi-Tenant Apps

AWS highlights use cases such as limiting AI agents and scoping down permissions based on runtime context. Before passing credentials to an agent, developers can use the API to restrict access to specific tasks, such as read access to a single Amazon Simple Storage Service bucket.

For multi-tenant applications using a single IAM role, a trusted credential broker can call GetScopedCredentials after authenticating a request to ensure workers can only access the files of the specific tenant who sent the request.

Our take

AWS has built a flexible primitive for workflows that need granular least privilege without spinning up entirely new IAM roles. While managing policy size limits and tracking policies via CloudTrail require careful engineering, the ability to apply up to three scope down policies gives teams robust runtime guardrails.

What to do

  • Define a scope down policy and call GetScopedCredentials using the latest AWS CLI or AWS SDK, ensuring your calling session has permissions for sts:GetScopedCredentials.
Key facts and where they come from
  • GetScopedCredentials is a new AWS STS API that scopes down existing temporary security credentials.
    GetScopedCredentials is a new AWS Security Token Service (AWS STS) API that you can use to scope down the permissions of existing temporary security credentials.
  • The API has a launch quota of 200 requests per second per account in each Region.
    The API has a dedicated request quota of 200 requests per second per account in each Region at launch.
  • A maximum of three scope down policies can apply to a session.
    You can apply a maximum of three scope down policies to a session…

The catch

  • The announcement does not say which specific AWS services are unsupported by GetScopedCredentials.
  • Generated policies using IAM Policy Autopilot may require manual review to replace wildcards with specific actions and resources.

The open question: How effectively will automated policy generators like IAM Policy Autopilot handle complex, dynamic permissions in multi-agent AI environments?

Tell us what you think · Discuss it in the Threat Intel room

Reader poll: Will you use GetScopedCredentials to restrict AI agent permissions in your AWS environment?

A show of hands from readers, not a scientific poll. Say why in the comments.

What can websites see about you? Run our free privacy check →

Add TechUpscale as a preferred source on Google

The Cyber Brief

Every weekday morning: what is being exploited, what to patch and who was hit, in one short email. Free.

I'm most interested in

How effectively will automated policy generators like IAM Policy Autopilot handle complex, dynamic permissions in multi-agent AI environments?

Comments are moderated. Be civil and stay on the story: no personal attacks, hate, spam or ads, and no one else's private details. Your email is never shown. Privacy policy.

More Cyber