Failed registering Scalable Target when defining auto scale option for ECS

Update

Mike Pope has published a nice article about Granting Permission to Launch EC2 Instances with IAM Roles (PassRole Permission) on the AWS Security Blog, which explains the subject matter from an AWS point of view.

Initial Answer

Skaperen's answer is partially correct (+1), but slightly imprecise/misleading as follows (the explanation seems a bit too complex for a comment, hence this separate answer):

To launch an EC2 instance with an IAM role requires administrative access to the IAM facility.

This is correct as such and points towards the underlying problem, but the required administrative permissions are rather limited, so the following conclusion ...

Because IAM roles grant permissions, there is clearly a security issue to be addressed. You would not want IAM roles being a means to allow permission escalation.

... is a bit misleading, insofar the potential security issue can be properly addressed. The subject matter is addressed in Granting Applications that Run on Amazon EC2 Instances Access to AWS Resources:

You can use IAM roles to manage credentials for applications that run on Amazon EC2 instances. When you use roles, you don't have to distribute AWS credentials to Amazon EC2 instances. Instead, you can create a role with the permissions that applications will need when they run on Amazon EC2 and make calls to other AWS resources. When developers launch an Amazon EC2 instance, they can specify the role you created to associate with the instance. Applications that run on the instance can then use the role credentials to sign requests.

Now, within the use case at hand the mentioned developers [that] launch an Amazon EC2 instance are in fact EC2 instances themselves, which appears to yield the catch 22 security issue Skaperen outlined. That's not really the case though as illustrated by the sample policy in section Permissions Required for Using Roles with Amazon EC2 :

{
   "Version": "2012-10-17",
   "Statement": [{
      "Effect":"Allow",
      "Action":"iam:PassRole",
      "Resource":"*"
    },
    {
      "Effect":"Allow",
      "Action":"iam:ListInstanceProfiles",
      "Resource":"*"
    },
    {
      "Effect":"Allow",
      "Action":"ec2:*",
      "Resource":"*"
    }]
}

So iam:PassRole is in fact the only IAM permission required, and while technically of administrative nature, this isn't that far reaching - of course, the sample policy above would still allow to escalate permissions by means of listing and in turn passing any available role, but this can be prevented by specifying only those roles that are desired/safe to pass for the use case at hand - this is outlined in section Restricting Which Roles Can Be Passed to Amazon EC2 Instances (Using PassRole):

You can use the PassRole permission to prevent users from passing a role to Amazon EC2 that has more permissions than the user has already been granted, and then running applications under the elevated privileges for that role. In the role policy, allow the PassRole action and specify a resource (such as arn:aws:iam::111122223333:role/ec2Roles/*) to indicate that only a specific role or set of roles can be passed to an Amazon EC2 instance.

The respective sample policy illustrates exactly matches the use case at hand, i.e. grants permission to launch an instance with a role by using the Amazon EC2 API:

{
  "Version": "2012-10-17",
  "Statement": [{
      "Effect":"Allow",
      "Action":"ec2:RunInstances",
      "Resource":"*"
    },
    {
      "Effect":"Allow",
      "Action":"iam:PassRole",
      "Resource":"arn:aws:iam::123456789012:role/Get-pics"
    }]
}

Scenario:

• AccountA - your account
• AccountB - customer account
• RoleA - Your account role to access DynamoDB, call STS, etc.
• RoleB - TrustedSecurityAuditor role in AccountB

You will need to manage two sets of credentials. Credentials A which are created from the IAM role assigned to the EC2 instance. Use these credentials to access your resources such as DynamoDB. Credentials B which are created via STS from RoleB. Use these credentials to access your customers resources.

You will then need to independently use the credentials based upon what you want to do. For example if you are using Python you would create two boto3 clients with different credentials.