Information security for cloud delivery
NCA registration
Registered cybersecurity service provider

ISO/IEC 27001
ISO/IEC 27001 certified
First steps
Four things, in this order
The whole sequence takes under an hour on a single resource. Doing it in this order costs nothing and preserves the ability to answer the questions that come next.
1 · Capture the access record first
Export whatever access logging exists for the resource before changing the permission — object-level access logs, the provider's audit trail, any CDN or gateway log in front of it. Some of these have short retention and some stop being queryable once the resource configuration changes. The evidence that answers "did anyone read it" still exists.
2 · Close it
Now make it private. Record the exact time you did, because the exposure window runs from whenever the permission was set to that moment, and the first of those two dates comes from the audit trail rather than from memory. The exposure ends, with both ends of the window dated.
3 · Establish what was actually in it
Not what it was for — what it contained during the window. A bucket described as holding marketing assets frequently also holds an export somebody put there once. Enumerate it rather than characterising it. A list, not an assumption, of what a reader could have taken.
4 · Find out how many others are the same
The configuration that produced this one was almost certainly applied elsewhere — the same template, the same script, the same person following the same runbook. Search the estate for the pattern before closing the incident. The incident covers the class rather than the instance.
The hard question
When the logs cannot tell you
Object-level access logging is not on by default on most providers, because it costs money and produces volume. So the common situation after finding an exposed resource is that the provider's audit trail shows when the permission was changed and by whom, and nothing shows whether anybody outside subsequently read the contents. That is an uncomfortable position and it is a very ordinary one.
What must not happen next is treating the absence of a record as evidence that nothing was read. It is not, and stating it that way in an incident report is the kind of sentence that is quoted back later. The honest formulation is that the resource was reachable for a stated period, that the contents were such and such, and that the access record needed to determine whether it was retrieved was not being collected.
There are partial signals worth checking even so. Egress volume on the account during the window can show a transfer far larger than normal traffic explains. Search engines and public code repositories sometimes hold a cached copy that dates the exposure differently from your own record. And where a CDN or gateway sat in front of the resource, its logs may cover what the object-level logging did not. None of these is conclusive, and the report should say which of them were checked.
How it happened
The four that account for most of these
- 01
Opened for a migration and never closed
A permission widened at eleven at night to unblock a data transfer, with every intention of reverting it in the morning. The most common single cause, and the reason temporary changes need an expiry rather than a note.
- 02
A template applied across an estate
One permissive default in a module reused by thirty deployments. The finding is never one resource, and searching for the pattern is what turns a small incident into a correctly sized one.
- 03
Public and private confused at the boundary
A resource genuinely intended to be public — a website's images — sharing a container or a policy with something that was not. The permission is correct for one half of what it covers.
- 04
A snapshot or backup shared outward
The database is locked down and a snapshot of it is shared with another account, sometimes an account that belonged to a vendor engagement that ended. Snapshots inherit none of the network controls of what they were taken from.
Where we help
What we can establish
What we do
- Triage the exposure the same day, starting with preserving whatever access record exists.
- Search the estate for the same configuration pattern, because the reported resource is rarely the only one.
- Put the preventive control where the cause was — usually in the deployment pipeline rather than in a policy document.
Where personal data was in the exposed resource, notification obligations and their timing are set by the data protection regulator and start from awareness rather than from confirmation. Get the technical scope to the people who make that determination on the first day.
Primary sourceCommon questions
What buyers ask
How do we work out how long it was exposed?+
From the provider's audit trail, which records configuration changes and normally retains them far longer than object access logs. Find the event that set the permission and the event that removed it; the window is between them. Where the audit trail predates your retention setting the answer may be "at least since the earliest record we hold", and that is the wording to use rather than a guess. Infrastructure-as-code history gives a second, independent date where the resource was deployed that way.
Should we turn on access logging everywhere now?+
On anything holding personal or commercially sensitive data, yes — the cost is small relative to being unable to answer this question, and this incident is the argument that gets it funded. Everywhere is usually not worth it: logging every read of a public image bundle produces volume that obscures the logs you would actually query. Decide by what the resource contains rather than by resource count.
What stops this happening again?+
A preventive control in the deployment pipeline rather than a detective one on the estate. A policy that refuses to deploy a publicly readable resource unless it carries an explicit exception tag catches the template case, the copy-paste case and the late-night case, before anything is created. Posture scanning still has a place for what is already there and for changes made outside the pipeline, but it tells you afterwards, and afterwards is where this incident came from.
In this cluster
Get same-day exposure triage
A same-day triage that establishes what was exposed, for how long, and whether the access record can answer who reached it.
What happens next
- 1We reply within one business day.
- 2A 30-minute scoping call with a senior consultant.
- 3A written proposal with scope, team, and timeline.
Or reach us directly
