Transcript
Summer, 2019. An attacker walked away from Capital One with over 100 million customer records. No password was cracked. Amazon's cloud was simply asked, and it said yes. Every request to AWS meets one checkpoint: IAM, short for Identity and Access Management. Think of each request as a form. Who is asking: here, a server role called web-firewall. What it wants: to read a file. Which resource: a storage bucket full of customer data. Plus context, like the IP address and the time. The answers live in policies, short rule cards. Each card says Allow or Deny, names actions and resources, and can add a condition. Rule one: the answer starts as no. If no card matches the request, it's denied. That's called an implicit deny. Rule two: IAM looks for a matching Deny first. If even one Deny matches, the answer is no. It doesn't matter how many cards say Allow. An explicit deny beats every allow. Rule three: only then does it look for an Allow. One matching Allow, and the request goes through. Now back to 2019. The attacker tricked a misconfigured firewall into sending requests as itself. IAM checked the form. No Deny matched. And its Allow reached over 700 storage buckets. Every check passed. IAM did exactly what the cards said. So the fix isn't a smarter checkpoint. It's better cards. Give each role only what its job needs: one action, on one bucket. Then add a guardrail Deny: no reading customer data from outside our own network. No Allow anywhere can override it. So remember it like this: start at no, Deny wins, and Allow only what's needed. Want to practise IAM policies hands-on? Try it free at seclern.com, and follow for more.