An AWS organization is easy to start and hard to restructure. The account layout, root user protection and audit log destination are decided in the first week, usually by whoever needed a sandbox that day.
These rules cover that first week and what follows, including Infrastructure as Code, identity, exposed credentials, Bedrock data residency and cost. Many assume a company that runs Microsoft Entra ID and Azure and is adding AWS alongside, and one rule maps each Azure service to its AWS equivalent.
Every AWS account has a root user. It can do anything in the account, including close it, and no IAM policy can restrict it. Too often it was created with one person's email address, a password they remember and an authenticator app on their phone. When that person leaves or loses the phone, you are locked out. If an attacker phishes them, the attacker owns the account.
An IAM user is a separate username and password in one AWS account, usually with an access key that sits in a developer's home folder for years. Your identity provider knows nothing about it, so your MFA and Conditional Access policies do not apply. When someone leaves, somebody has to remember to delete their IAM user in every account.
Most teams start on AWS with one account that holds production, test servers, audit logs and the billing owner. Then a test server is compromised, or a cleanup script hits the wrong environment, and nothing stands between the mistake and production.
The first AWS account is usually built by clicking through the console, with a promise to "script it later". Six months on, nobody can say why a Prod security group is open to the internet, NonProd has settings nobody chose, and the person who clicked has moved to another project.
After a security incident, the first question is "what happened?". The audit log only answers it if the log is complete and nobody could have changed it. If the logs sit in the workload account, an attacker with admin access there can stop the trail and delete the evidence. A SOC 2 or ISO 27001 auditor asks the same thing: how do you know these logs were not altered?
An audit log tells you what happened, but only when someone reads it. If nobody is told when the root user signs in or when logging stops, you find out weeks later at the next audit, or halfway through an incident.
An access key gets pushed to a public repository, pasted into a support ticket or left on a stolen laptop. The usual reactions make things worse. Someone rewrites the Git history and tells nobody, so the key still works. Or someone deletes the key and stops there, while the attacker's sessions keep running and the user they created stays behind.
IAM policies decide what each person or role can do, but an account administrator can rewrite them. If an IAM policy is the only thing keeping your data in Australia or keeping CloudTrail on, any admin in that account can remove it.
The management account owns your AWS organization. Whoever controls it can create and close accounts, rewrite every guardrail and pull any account out of the organization. Many teams still treat it as an ordinary account: the first workload was deployed there, everyone signs in daily, and it has the same SSO access as everything else.
AWS charges for what you use and invoices after the month ends. A test loop that calls an AI model all weekend, or a forgotten GPU instance, can run up thousands of dollars before the invoice shows it. It gets worse in a company that mostly runs on Azure, where the AWS bill goes to finance and no engineer ever opens it.
AWS never deletes anything for you. The load balancer from last year's proof of concept, the NAT gateway in a VPC nobody uses and the IAM role for a contractor who left are all still there. Each one costs money or gives an attacker a way in. Nobody removes them, because nobody is sure that nothing depends on them.
Plenty of organizations run almost everything on Azure and use AWS for one thing, such as Amazon Bedrock. The identity team audits Entra ID every quarter, but AWS IAM is a second system that one or two people understand. A server that needs AWS access ends up with an IAM access key that nobody on the Azure side can see.
Workloads that run in AWS get credentials from an IAM role. A server in your own data center has no role, so the usual workaround is an IAM user whose access key is copied onto the server. That key never expires, works from any network and is rarely rotated. Anyone who can read the file can use it.
A database password ends up in a Lambda environment variable, a Terraform variables file or an ECS task definition. Everyone who can open the function's configuration can read it. It is in Git history for good, and changing it means a redeploy, so nobody ever changes it.
A team creates one role called lambda-shared-role for its first function. Each new function reuses it, and each one adds the permissions it needs. A year later the function that resizes thumbnails can also read the orders table and send email. One vulnerable package in any function now gives an attacker everything that all of them can do.
Teams with data residency obligations often pick Amazon Bedrock because AWS has regions in their country. Then a developer copies a model ID from a sample, the ID starts with global., and prompts from a Sydney workload get processed in Ohio or Frankfurt. The call still goes to the Sydney endpoint and CloudTrail still logs it in Sydney, so nobody notices.
If your security team works in Microsoft Sentinel, AWS activity that never reaches it is a blind spot. Watching it means a second console, query language and set of alerts, so in practice nobody looks until something has gone wrong.
A team that knows Azure opens the AWS console and finds hundreds of services with names like Route 53, Fargate and Athena. Searching for the Azure name finds nothing. Worse, a few names do exist in both clouds and mean different things, so people assume they know how something works and get caught.