A new cloud account arrives in its least secure state. There is one all powerful credential, no audit configuration, no spending guardrail and no record of who did what. Everything after that is either an improvement or an accumulation of debt. The work below takes well under an hour and is far cheaper now than after the account is in use.
1. Take control of the top level credential
Every provider has an account level identity that can do anything, including closing the account and changing billing. Whatever it is called, treat it as a break glass credential rather than a login.
- Change the password to one generated at random and stored in a password manager or secrets vault.
- Enable multi-factor authentication on it immediately.
- Confirm the recovery email and phone number point at something the organisation controls, not an individual’s personal address.
- Write down who has access to it and under what circumstances it is used.
If the account was obtained rather than created, rotating this credential is not optional. Until it is rotated, you do not know who else can use it.
2. Create a working identity and stop using the top level one
Create a separate administrative user or role for day to day work, with multi-factor authentication of its own, and use that. The top level credential should then go unused for long stretches, which is exactly the point: any sign of it being used becomes a signal worth investigating.
3. Remove what you did not create
In an account with any prior history, enumerate and review existing identities, access keys, roles and any third party integrations. Delete or disable anything that is not accounted for. Pay particular attention to long lived access keys, which are the credential type most likely to have leaked, and to any identity that can modify permissions.
4. Turn on the audit trail before you need it
Every provider offers an activity or audit log that records control plane actions. It is frequently off, partially configured, or retained for only a short window by default. Enable it across all regions, send it somewhere the account’s own administrators cannot quietly delete, and set a retention period that matches how long you would plausibly need to answer a question about the past.
5. Put a spending guardrail in place
A compromised cloud account is usually monetised by running expensive compute, and the first sign is the bill. Set a budget with alerts at thresholds that would be abnormal for your workload. This is a detection mechanism as much as a financial one.
6. Apply least privilege where it costs nothing
Full least privilege is an ongoing discipline, but the cheapest portion is available immediately. Grant permissions to groups or roles rather than individuals. Scope automation credentials to the specific actions they perform. Prefer short lived credentials over static keys wherever the platform offers them. And review anything that grants permission to change permissions, because that is the one privilege that undoes all the others.
What this buys you
None of these steps prevent every incident. Together they ensure that if something does go wrong, access is limited, the event is recorded, the bill is capped and the credential that could do the most damage is not in daily circulation. That is most of the practical difference between a contained incident and an open ended one.