
From az login to Bedrock: AWS Auth for Azure Switchers
If you're used to Azure AI Foundry's API keys, Bedrock's authentication model might feel unfamiliar at first — here's the mental model, and the two-command equivalent of az login.
Before we start: three quick terms
- IAM is AWS's identity and access system — the equivalent of Entra ID. Every request to any AWS service is checked against it.
- SigV4 is the signing method AWS credentials use to prove a request is legitimate. Instead of attaching a static token to a request (like a password), the request itself gets cryptographically signed using your credentials. More on why that matters below.
- IAM Identity Center is AWS's answer to "log in once, get access to what you're allowed to use" — the rough equivalent of your organization's Entra ID tenant. If your organization uses it, this is what you'll authenticate against.
The question that started this
If you've used Azure AI Foundry, you've probably seen the setting that
disables API key authentication for a deployed model, forcing every
caller to use Entra ID instead. It's a genuinely good security
practice — and if you're moving from Azure to AWS Bedrock, a natural
first question is: where's the equivalent setting for Bedrock?
Short answer: there isn't one to find, because there's nothing to turn
off. Bedrock's native, original authentication method has always been
IAM credentials — the AWS equivalent of Entra ID — not an API key.
Keys were added later as an optional convenience feature. If you
never opt into using one, you were never authenticating with a key in
the first place.
Why this is confusing coming from Azure
Azure OpenAI and AI Foundry ship with both an API key and Entra ID
authentication turned on simultaneously. You have to actively go find
the setting and switch keys off. That's the model most people learn
first, so it's reasonable to expect AWS to work the same way — a
toggle somewhere in the Bedrock console labeled "disable API keys."
AWS built it the other way around. IAM has always been the only way in.
API keys, when AWS introduced them, were explicitly framed as a
getting-started convenience — useful if you're following a quick
tutorial and don't want to learn IAM roles and SigV4 signing on day
one. If you go through the normal AWS login flow instead, you never
touch that feature at all.
The actual equivalent of az login
With Azure, you'd typically run:
az loginand then call Azure AI Foundry using whatever identity that logged you
in as. The AWS equivalent, if your organization uses IAM Identity
Center (common in larger orgs):
aws configure sso # one-time setup: point this at your org's IAM Identity Center URLaws sso login --profile your-profileThis opens a browser, you authenticate the same way you would for any
other company login (possibly including MFA, depending on how your
org has it set up), and AWS caches short-lived credentials locally.
From there, calling Bedrock from code looks like this:
import boto3
client = boto3.client("bedrock-runtime", profile_name="your-profile")response = client.converse( modelId="anthropic.claude-sonnet-4-5-20250929-v1:0", messages=[{"role": "user", "content": [{"text": "Hello!"}]}])No key anywhere in that code. The SDK automatically finds your cached
login credentials and signs each request for you.
One real difference worth knowing (not just naming)
Here's a genuine trade-off, not just a vocabulary difference. With
Azure, once you have a token, you can paste it into a raw
curl
command as a header and it works:curl -H "Authorization: Bearer <token>" https://your-endpoint...You can't do that with plain IAM credentials on AWS. SigV4 requires
computing a signature from parts of the request itself — you can't
just stick a static value in a header and call it done. This is
exactly why AWS eventually added the API key feature: it gives people
who want "paste a token, curl it" simplicity the option to have that,
at the cost of stepping outside AWS's stronger default security model.
So the honest comparison isn't "Azure is less secure by default" — a
default Azure OpenAI resource genuinely does have both key and Entra ID
auth enabled at once, so on day one it is more permissive than a
default Bedrock setup. The interesting difference is direction: Azure
starts permissive and asks you to lock it down; AWS starts locked down
and offers you a permissive option if you want it.
What to take away
If you're a student learning cloud AI services: the practical
takeaway is simple — if you're setting up Bedrock for a project or
class, reach for
aws configure sso / aws sso login (or a plain IAM
role if you're on a personal account without Identity Center) before
reaching for an API key. You'll get the same working setup, without
picking up a habit that's harder to unwind later.If you're already working with this professionally: if your
organization already federates IAM Identity Center to Entra ID (a
common setup when a company already runs Microsoft 365 or Entra ID
elsewhere), logging into AWS this way authenticates against the exact
same identity, MFA, and Conditional Access policies your team already
uses for Azure. It's not a second identity system to manage — it's
the same one, with a different login command.
If you do need to allow API keys in your organization — or want to
lock them down explicitly rather than relying on nobody using them —
that's a deeper, policy-level topic on its own.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article