A dark indoor swimming pool lit by cool blue reflections, with a metal handrail and a no diving sign at the water's edge

The one where we signed up to the admin pool.

One cloud misconfiguration hid among 2,500 triggered rules. An identity pool that let anyone register an account ultimately reached 34,000 production records.

Across a client’s cloud estate, 2,500 rules were triggered due to potentially misconfigured resources. Eight were medium severity, and all eight related to the same identity service. Hidden among them was an identity pool that allowed anyone to register a new account.

That account could ultimately read, change and delete 34,000 records spanning multiple production databases. 

On its own, that setting does not necessarily constitute a misconfiguration. Any application with a user registration flow needs it enabled. Here, it was a lower-severity misconfiguration that could easily be overlooked among thousands of potential findings. 

Medium severity. Easy to miss. Not harmless. 

The needle 

Our operator inspected the identity pool and found only internal administrative users inside.

That was the first red flag because the configuration didn’t seem intended. 

The pool also did no domain validation and used no client secrets, which are passwords an app must present to prove it’s the real app.

Both are checks that stop strangers from signing up. 

Two values, one web address 

An attacker needs only two things to register: the client ID (a public label for the app) and the cloud region.

The cloud provider treats both as non-secret, assuming other controls are in place, such as disabled self-registration, client secrets, or domain validation. 

Those values also must sit in an app’s front-end config files for sign-up to work. Here, the login pools were served through externally reachable content delivery distributions, and one held a config.js file with both values. Anyone who knew the address could register an account in the admin pool. The same file hinted at the APIs a new account could call. 

Registering took a web address and two values from a public file.

Further research showed that any registered user could call those APIs, which in turn triggered serverless functions (small pieces of code that run on demand). Those functions gave full read, write and delete access to 34,000 records in various production databases. 

Medium for whom? 

A scanner rates a setting by how it usually behaves. An attacker rates it by what it opens. In our view, that gap is where the real risk sits. 

Access to the client’s cloud accounts helped our operator confirm the exploitation path faster. But this is a common misconfiguration, and testing for it from a closed-box view needs only a few details that were meant to be public. 

Our operator also improved our internal recon tooling, so the whole Swarm is now better at spotting this pattern in future work. 

Unleash the Swarm and find out which of your medium findings is really a way in.