Effective cloud threat detection SIEM programs start from an uncomfortable observation: most correlation rules in use today were written for networks, not for control planes. They look for a workstation touching many servers, a service account logging on in unusual places, or a burst of failed authentications. Cloud reconnaissance does not look like that. It looks like a valid identity making valid API calls, only in an order and at a pace that no administrator would use.
This article is written for security operations managers and SOC leads at Saudi enterprises that are adopting SaaS platforms and running hybrid infrastructure. It describes four behavioral indicators that rules commonly miss in Microsoft 365, Azure and AWS, and three baselines that help separate legitimate administration from credential compromise.
Why on-premises rules fall short in the cloud
On-premises lateral movement leaves network artifacts: connections between hosts, remote execution, authentication events on domain controllers. In cloud and SaaS environments the attacker rarely needs to move between hosts at all. A single stolen token or access key can read configuration, list resources and query data through the provider's own APIs.
Two consequences follow. First, the events that matter are control-plane and audit-log events, which are often ingested but not correlated. Second, the individual events are usually benign. A call to list users or describe roles is routine. The signal is in the sequence, the source and the context.
Four behavioral indicators that rules often miss
1. API enumeration across services
Administrators query what they need. Attackers querying with a new identity tend to ask broad questions: list all users, list all roles, list all storage locations, list all subscriptions or accounts. Look for a single identity calling many distinct read or list operations across several services in a short window, especially when that identity has no history of doing so. Volume alone is a weak signal; breadth of distinct operations is stronger.
2. Permission probing and escalation attempts
Early-stage actors test what they are allowed to do. This shows up as repeated access-denied responses followed by a success on a related action, attempts to read policy or role-assignment details, or attempts to create credentials, consent grants or role assignments. A failed privileged action from an account that never attempts privileged actions deserves attention even when nothing succeeded.
3. Cross-tenant and cross-account queries
In multi-tenant and multi-account setups, look for an identity that begins to touch tenants, subscriptions, accounts or mailboxes it has never accessed. Guest accounts, delegated administration and federated roles are legitimate paths, which is why they are also useful to an attacker. The indicator is a first-time relationship combined with data-oriented queries rather than configuration changes.
4. Identity context that does not fit the session
Reconnaissance is often preceded by a sign-in that passes multifactor checks but differs from the user's pattern: a new user agent, an automation-style client used by a person, a token reused from a different network, or a session that moves directly to directory and role queries without touching the applications the user normally uses. Rules that evaluate sign-in risk and subsequent activity separately miss the link between them.
Three baselines that separate administration from compromise
Detection logic improves when it compares activity with a defined normal rather than with a fixed threshold. Three baselines carry most of the value.
Baseline 1: Role-and-task baseline
Document which administrative roles perform which tasks, and from which management paths. A help-desk role resetting passwords is expected; the same role enumerating role assignments is not. Build rules from the role's expected operations, and flag operations outside that set.
Baseline 2: Time-and-change-window baseline
Legitimate administration is usually tied to change records, maintenance windows and known working hours. Correlate privileged activity with approved changes where you can. Activity with no matching change, particularly outside working patterns in the Kingdom, should score higher. Allow for known exceptions such as on-call rotations, and record them rather than silencing the rule.
Baseline 3: Identity-and-source baseline
For each privileged identity, establish usual sources, client types, tenants and accounts. Distinguish human administrators from service principals and automation, since their behavior differs sharply. A service principal that starts making interactive-style discovery calls, or a human administrator using a pattern typical of scripts, is a stronger signal than either event alone.
Turning the indicators into detection logic
A practical approach is to score combinations instead of alerting on single events:
- Choose the identities that matter most: global and subscription administrators, privileged service principals, and accounts with delegated access.
- Record the baseline for each class of identity using the three baselines above.
- Write rules that combine an identity-context anomaly with broad enumeration or permission probing within a defined window.
- Test each rule against known administrative workflows, so that migrations, audits and onboarding work do not generate constant noise.
- Review results regularly and retire or tune rules that analysts routinely close without action.
Also confirm that the logs you depend on are collected, retained and complete. A rule cannot fire on audit data that was never enabled.
When an organization needs outside support
The signs that it is time to review your cloud detection are usually practical:
- Cloud and SaaS logs are ingested, but few correlation rules use them.
- Analysts cannot say which detections cover reconnaissance rather than only known malware or failed logons.
- Administrative activity is noisy, and alerts are closed without investigation.
- Managed service contracts were signed before cloud adoption expanded, and nobody has checked them against current requirements.
- Incident response plans have not been rehearsed against a cloud scenario.
What a good outcome looks like
A well-run security operations capability can show which cloud threats it detects, which it does not, and why. Detection use cases are prioritized by threat and by regulatory obligation, so effort goes where exposure is highest. Plans are rehearsed, and reporting is ready to satisfy notification rules from the relevant regulator, such as the National Cybersecurity Authority or SAMA for institutions under its supervision, and the data-protection authority where personal data is involved.
How we approach it
Our Security Operations Advisory service covers the design, assessment and improvement of security operations, monitoring and incident response capabilities. The aim is detection and response that meets regulatory expectations.
The engagement runs 4–10 weeks depending on scope. The deliverables are:
- A security operations maturity assessment
- A detection use-case catalog
- An incident response plan and playbooks
- A tabletop exercise and lessons learned
- An improvement roadmap
In practice, this gives you a measured view of your detection and response maturity, with detection use cases prioritized by threat and obligation. We also review managed service contracts against requirements, so that outsourced monitoring is held to what your organization needs.
A common question is whether we run a managed SOC. Our role is advisory: we design, assess and improve security operations and help you select and oversee providers, so our advice stays independent.
Next step
If your SIEM was built around on-premises behavior and your estate now spans Microsoft 365, Azure and AWS, a structured review of detection coverage is a sensible first step. Speak with CISO Consulting about Security Operations Advisory to establish where your cloud reconnaissance coverage stands and what to improve first.