Security
Built on the assumption that a key will leak.
We have responded to a real cloud compromise. It changed how we set up every account. This page describes the baseline we run for ourselves and for the clients whose infrastructure we manage.
The baseline
Six layers, each one assuming the others can fail.
Identity
- Single sign-on for every person. No stored access keys on laptops, in .env files or in CI.
- Credentials expire within a working day and are re-issued on sign-in.
- A documented break-glass path for emergencies, tested before it is needed.
Guardrails
- Organisation-level policies sit above every account, including administrators.
- Activity is limited to the regions the business uses.
- Services that attackers abuse for cost, such as AI model hosting and marketplace purchases, are denied wherever the business does not use them.
Audit
- One audit trail covers every account and every region, with log file validation.
- Logs are written to a separate account. Workload accounts cannot stop or alter the trail.
- Application-level audit history: who changed which record, and when.
Data
- Production, staging and management run in separate accounts.
- Databases are private. Access goes through short-lived, audited tunnels with throwaway keys.
- Secrets live in a secrets manager, never in the repository.
- Multi-tenant systems scope every query by organisation, and guard tests fail the build if one does not.
Detection
- Budgets per account with alerts on actual and forecast spend.
- Daily cost anomaly alerts that reach a person, not a dashboard.
- Health checks on every production service.
Delivery
- Linting, type checks and automated tests run before anything is built or deployed.
- Infrastructure is defined in code and reviewed like application code.
- Business rules are enforced on the server and covered by tests, not just hidden in the interface.
Why we are strict about this
The fastest way to a large bill is one leaked key.
Stolen cloud credentials are now routinely used to run AI models on the victim's account. With the right guardrail in place, the same stolen key gets an access-denied error instead.
- We audit existing AWS accounts and report exactly what each credential can reach.
- We put the baseline in place, usually within a few days.
- If something has already happened, we contain it and find what was left behind.
Confidentiality and data protection
We sign non-disclosure agreements before discussing project details. Client data is processed only for the purpose of the engagement, in the cloud region the client chooses, and we tell you which third-party services, including AI model providers, will see which data before any of it is sent.
Product illustrations on this website use invented sample data. Screenshots come from live systems with names and other personal details obscured. We do not publish client records.
Start a conversation
Worried about what is in your AWS account?
Ask for an access review. We will tell you what every key, user and role can do, and what to fix first.