Skip to content

Blueprint, draft v1.0

Atarus Offensive Cloud Professional: exam blueprint

This blueprint says what the AOCP exam tests, how it is run, what you may and may not do in it, and how a pass is decided. It is the document the exam is written against: every exam objective traces back to an objective here, and every objective here traces back to a free page in the Atarus Learn library, named next to it.

Draft v1.0. The final blueprint is in development. It is published after every Cloud page it cites carries captured output from a real account. Until then the objectives below may still change, and the change log at the end records what did.

Purpose and who it is for

The exam is for people who test cloud accounts for a living or want to: penetration testers moving into cloud work, cloud and platform engineers who review their own accounts, and detection engineers who want to know what an assessment looks like from the other side. You should be comfortable with a command line and the AWS CLI, and able to read an IAM policy document. You do not need the paid course to sit the exam. Everything the exam rests on is on the public Learn pages cited below.

What the credential attests

A holder of the AOCP sat an eight-hour practical exam in a real AWS account issued by Atarus Offensive Security LLC, completed at least three of four flagged objectives without changing anything the objectives did not ask for and without printing a secret value, delivered a report that an engineer could reproduce every finding from, including what CloudTrail showed for each step they took, with every finding backed by their own evidence and the lab's logs. It says nothing more than that. It is a statement by the issuer about one exam on one date, and the verifier page shows which blueprint version the holder sat. The name on a credential is the name the candidate gave. We do not check identity documents.

Exam format

Where
A leased AWS account, built for your sitting and wiped after it. AWS only in version 1.
How long
Eight hours of lab time. Your credentials are a role session that expires when the lease ends, never later.
What you do
Four flagged objectives plus the report. Each objective has a flag you submit and an evidence requirement you meet in the report.
The report
You have 48 hours after the lab window closes to deliver it. Every finding carries reproduction, evidence location, severity reasoning, the fix and the defender's view.
Grading
Against the published rubric, with the help of an AI tool. A person confirms every result, and you can ask for a human re-grade.
AI assistants
Allowed in the lab window and for the report, if your report declares each one and what you used it for. See the AI policy below.
When
Weekday sittings. The start times are published with the final blueprint.
Your layout is your own
Each sitting is generated from one of three templates. Names, account ids, role names, bucket names, parameter paths and flag values differ per sitting, so no two candidates see the same strings.

The objectives in a set follow this shape: from a CI deploy key, map the identity and reach a second role through a trust policy misconfiguration; find the secret names exposed in function configuration and user data across two regions without printing a value; identify the guardrail gap that lets the escalation go unseen and show what the defender would have seen; and prove impact without damage by reading one named object and one named record.

There is no webcam, screen or room recording and no identity check. The lab's logs and your own evidence carry the exam, as the standards page explains.

Domains and objectives

Domain weights show how the exam's objectives and report items are spread across the content. The weights are a draft until the final blueprint. The score itself comes from the rubric on the standards page. Each objective is something you do and a grader can see in your evidence or your telemetry. The pages named under each objective are the Learn pages it rests on.

20 percent, draft weight

Domain 1. Identity enumeration and access mapping

  1. Objective 1.1. Confirm the identity behind the credentials you were given, the account it lives in and whether it is a user, a role session or a CI principal, before any other call.

    Rests on cloud-account-first-hour

  2. Objective 1.2. Enumerate every policy that applies to the identity: attached directly, inline, through groups, through roles it can assume, and any permissions boundary on it.

    Rests on cloud-account-first-hour aws-iam-privilege-escalation-checks

  3. Objective 1.3. Map every role in the account to what its trust policy allows, and mark each trust that names the account root, names *, or crosses accounts without an sts:ExternalId or aws:PrincipalOrgID condition.

    Rests on aws-iam-privilege-escalation-checks cloud-persistence-audit

  4. Objective 1.4. Identify how a deploy credential authenticates (a stored long-lived key or OIDC federation) and, for any web identity trust, record the exact subject and audience condition that limits it.

    Rests on ci-pipeline-oidc-trust-review

  5. Objective 1.5. Record every principal in the account that holds administrator rights and whether that access is standing or time-bound, reporting holders by account type rather than by person.

    Rests on identity-privileged-role-pim-review cloud-account-first-hour

25 percent, draft weight

Domain 2. Privilege escalation paths

  1. Objective 2.1. Determine with the IAM policy simulator whether the identity can rewrite its own permissions (new policy versions, a permissive older default version, policy attachment, access keys on other users), and read every non-default version of the policies attached to it.

    Rests on aws-iam-privilege-escalation-checks

  2. Objective 2.2. Identify pass-role paths: iam:PassRole together with permission to create or update Lambda functions or EC2 instances, and lambda:UpdateFunctionCode on a function whose role is stronger than your own.

    Rests on aws-iam-privilege-escalation-checks cloud-account-first-hour

  3. Objective 2.3. Assume a role reached through a trust policy misconfiguration and prove it by printing the session ARN and expiry, never the keys.

    Rests on aws-iam-privilege-escalation-checks

  4. Objective 2.4. Identify paths to an instance profile's role through SSM or EC2 actions, and record each instance whose metadata service answers without a session token (HttpTokens of optional).

    Rests on aws-iam-privilege-escalation-checks

  5. Objective 2.5. Read what each reached role can do, follow a chain of two roles inside the leased account, and stop at the step the objective asks for.

    Rests on aws-iam-privilege-escalation-checks ci-pipeline-oidc-trust-review

  6. Objective 2.6. For each path, record the policy statement that grants it, quoted as the API returned it with its Sid and, for a managed policy, its version id.

    Rests on aws-iam-privilege-escalation-checks cloud-finding-report

  7. Objective 2.7. Where an objective asks for a demonstration that changes something, make the least-impact proof on a resource the objective names and record the command that reversed it. Where the objective does not ask, a simulation result is the proof.

    Rests on aws-iam-privilege-escalation-checks cloud-impact-without-damage

15 percent, draft weight

Domain 3. Secrets, storage and data exposure

  1. Objective 3.1. Find secrets in Lambda function configuration, EC2 user data and Parameter Store across every region in scope, printing variable names, line numbers and parameter types, never values.

    Rests on cloud-account-first-hour aws-secrets-kms-key-policy-review

  2. Objective 3.2. List every Secrets Manager secret with its rotation state and its key, and every parameter that holds a credential as a plain String.

    Rests on aws-secrets-kms-key-policy-review

  3. Objective 3.3. Determine who can read and decrypt a named secret by reading all four places that decide it: the identity policy, the secret's resource policy, the KMS key policy and any grant on the key.

    Rests on aws-secrets-kms-key-policy-review

  4. Objective 3.4. Identify KMS key policy statements with a * principal and no organization or account condition, and grants that the account's own layout does not explain.

    Rests on aws-secrets-kms-key-policy-review

  5. Objective 3.5. Check S3 exposure at account and bucket level (the public access block, the bucket policy, anonymous listing) and, where a bucket answers, keep the proof to one listing of at most three names or the headers of one object.

    Rests on storage-bucket-exposure-checks cloud-account-first-hour

  6. Objective 3.6. Check the other public exposure an account allows: snapshot public sharing state in each region in use, and function URLs with an AuthType of NONE.

    Rests on cloud-account-first-hour

15 percent, draft weight

Domain 4. Guardrails, logging and the defender's view

  1. Objective 4.1. Determine whether CloudTrail is logging in every region in use, using the trail status rather than the trail's existence, and whether the trail is multi-region and excludes KMS or data events.

    Rests on cloud-account-first-hour aws-secrets-kms-key-policy-review

  2. Objective 4.2. Identify the guardrail gap that lets the escalation go unseen, either a trail setting or a service control policy that does not deny what the path uses, and state the deny statement or trail setting that would close it.

    Rests on aws-organizations-scp-review cloud-account-first-hour

  3. Objective 4.3. For each step you took, name the CloudTrail event it produced, its event source and the region it lands in (IAM and global STS events land in us-east-1), and find your own events in the account's event history.

    Rests on aws-iam-privilege-escalation-checks aws-organizations-scp-review cloud-impact-without-damage aws-cloudtrail-guardduty-evidence-review

  4. Objective 4.4. State which of your events should have raised an alert and why, for example a policy simulation from a CI key, or a launch dry run from a key that never launches compute.

    Rests on aws-iam-privilege-escalation-checks cloud-impact-without-damage ci-pipeline-oidc-trust-review

  5. Objective 4.5. Where a claim rests on an event the account does not log, such as an S3 data event or a KMS Decrypt the trail excludes, report the logging gap as a finding of its own.

    Rests on cloud-finding-report aws-secrets-kms-key-policy-review

10 percent, draft weight

Domain 5. Persistence and impact without damage

  1. Objective 5.1. Identify footholds created in a stated window (IAM users, access keys, changed role trust policies, new functions) each with the event that created it, the identity that made it and the time.

    Rests on cloud-persistence-audit

  2. Objective 5.2. Report a suspected foothold without removing it, so the evidence stays in place for the account owner.

    Rests on cloud-persistence-audit

  3. Objective 5.3. Prove read access to data with a count and one object: list and size what the identity reaches, read only the object or record the objective names, and keep its hash rather than its contents.

    Rests on cloud-impact-without-damage

  4. Objective 5.4. Show a destructive permission with the policy simulator instead of using it, and name any resource policy that would still stop it.

    Rests on cloud-impact-without-damage

  5. Objective 5.5. Where an objective asks for proof of write or cost impact, use one tag written and removed, or a dry run priced from the public pricing page and labelled as an estimate.

    Rests on cloud-impact-without-damage

  6. Objective 5.6. Reconcile every write event under your identity in CloudTrail against the changes you declared, and explain any other before you write anything else.

    Rests on cloud-impact-without-damage

15 percent, draft weight

Domain 6. Evidence and reporting

  1. Objective 6.1. Build an evidence package in which every claim is tied to the command that produced it and its output, with no secret value printed anywhere in it.

    Rests on cloud-finding-report writing-a-finding

  2. Objective 6.2. Name each affected resource by account, region, full ARN and unique id.

    Rests on cloud-finding-report

  3. Objective 6.3. Quote the statement or setting at fault as the API returned it, or the empty response where a setting is missing, and where no single statement is at fault, say which pair creates the path.

    Rests on cloud-finding-report

  4. Objective 6.4. Point each claim at its log event: the API call, the time in UTC, the account, the region, your identity and the event id.

    Rests on cloud-finding-report

  5. Objective 6.5. Write each finding with a title that states the impact, a severity whose reasoning matches what the evidence proves, and impact shown kept apart from impact assumed, in business terms. Whether a CVSS vector is required on each finding is published with the final blueprint.

    Rests on writing-a-finding

  6. Objective 6.6. Give each fix as the specific policy, setting or record, as a console path and as code, with the caveat to test it outside production first.

    Rests on cloud-finding-report writing-a-finding

  7. Objective 6.7. State what you did and did not review, and end each finding with a one-command retest and its expected result.

    Rests on cloud-finding-report writing-a-finding

In the course only, and why

The paid course covers more than the exam. These topics are taught and are not examined in version 1:

The reason is the same in each case: the exam only tests what we can run in a real environment, with enforced permissions and full telemetry, the same way for every candidate. Version 1 can do that on AWS. When another platform can be run that way, it enters the blueprint as a new version with a change log entry.

Allowed tools

You may use any tool you run from your own machine against the leased account: the AWS CLI and SDKs, open source enumeration tools such as Prowler and the firm's own atarus-cloud, your own scripts, and commercial tools you hold a licence for. Whatever the tool, the prohibited actions below apply to what it does, and a scanner that writes to the account or prints a secret value counts as you doing it. Your notes, the Learn library and public documentation are open throughout.

AI assistants must be declared in the report, as the AI policy below says. The full tool list, and whether the report must also declare the other tools you used, is published with the final blueprint.

AI policy

You may use AI assistants during the eight-hour lab window and to help write your report, as long as you declare them.

  • Declare each one. Your report has an AI declaration section that lists each assistant you used and what you used it for. If you used none, the section says so.
  • Your evidence carries every finding. Every finding must be backed by evidence from your own lab: the commands you ran, their output, and the lab's own logs, which CloudTrail writes to an account you cannot reach (not the trail inside your lab account). A claim the logs do not support fails that item, however well it is written.
  • What an assistant does counts as yours. The prohibited actions below apply to anything an assistant runs or writes for you. An action outside scope, or a secret value printed, counts as yours.
  • A false declaration can cost the credential. A declaration later shown to be false, such as an assistant you used and did not list, is grounds for revoking the credential under the sanctions on the standards page: revocation shown on the verifier, and a 12-month bar from our exams.

Your report is graded against the rubric with the help of an AI tool, and a person confirms every result. The standards page says what the tool does and does not do, and how to ask for a human re-grade.

Out of scope and prohibited actions

  • No denial of service. Nothing that floods, exhausts or degrades a service, a quota or the lab platform.
  • Nothing outside the leased account. Other accounts in the lab organization, the logging account, the lab platform, other candidates and any third party are out of scope.
  • No changes the objective does not ask for. Every objective is a read or a least-impact proof. Where an objective asks for a change, make that change and reverse it.
  • No secret values printed. Names, types, counts, line numbers and hashes only, in your terminal, your evidence and your report.
  • No social engineering, and no attempt to stop, alter or evade CloudTrail.
  • No sharing exam content during or after the sitting.

Any out-of-scope action that appears in the telemetry fails a must-pass report item, whatever your score.

The standard

The pass decision is conjunctive. You pass when all three hold:

  1. At least three of the four objectives are completed.
  2. Every must-pass report item is met: no secret values printed, every finding reproducible, and no out-of-scope action in the telemetry.
  3. Your total score is at or above the cut.

The cut is set before the beta sittings by the programme owner and three to five outside reviewers who do not sit the exam, then checked against the beta results. It is provisional for the first 50 sittings and labelled as such on the standards page, and it is reviewed when those sittings are done. The rubric, with the points for each part, is on the standards page.

Retake, validity and renewal

  • Retake. One free retake is included. The waiting period before it, the terms for attempts after it, and whether a retake draws a different template are published with the final blueprint.
  • Validity. The credential is current for three years from the date of issue. The verifier shows "current until" with the date, then "lapsed". The record is never deleted.
  • Renewal. A free renewal set of two hours with no report. When the renewal window opens, and for how long, is published with the final blueprint.

Accommodations and appeals

Accommodations such as extra time and breaks are agreed on request before you book, and they never appear on the credential. You can ask for a human re-grade at no cost: your report is graded again by hand, without the AI tool's suggestions. How to ask and the time limits are published with the final standards. You may also appeal a result within 14 days. A grader who did not grade your report reviews it, and you get a decision within 15 business days. The full process, the rubric, who grades and how the standard is set are on the standards page.

Accreditation statement

Atarus certifications are issued by Atarus Offensive Security LLC. They are not accredited by ANAB, ICE or any other accreditation body, and they are not approved under DoD 8140 or any government program. Our standards page shows exactly how each exam is built, graded and reviewed, and our pass rates are published every quarter. If our status changes, this page will say so, with the date and scope.

Version and change log

The blueprint is versioned. A change to a domain, a weight or an objective is a new version with an entry here, and the verifier shows which version each holder sat.

Blueprint versions
Version Date Change
1.0 draft 4 Nov 2026 AI policy set: AI assistants allowed in the lab window and for the report when the report declares them, with every finding resting on the candidate's evidence and the lab's logs. Grading with the help of an AI tool, a person confirming every result. The identity check and the debrief dropped. Start times to be published with the final blueprint.
1.0 draft 29 Oct 2026 First draft. Once the final v1.0 is published, every change after it is listed here.