- In short
- The do-not-ship-without-review categories are a standing list of output types that always require human review before release, decided in advance rather than case by case: final client deliverables, audit-critical or financially material calculations, anything involving regulated, confidential, or highly sensitive data, and public or legal communications where a misstatement carries lasting consequence. The list is fixed and not waived because an output looks polished.
A standing list, decided ahead of time
The four risk thresholds are a judgement you apply case by case. The do-not-ship-without-review categories are the opposite: a fixed list, settled in advance, of output types that always require human review before release. The Claude Certified Associate - Foundations (CCAO-F) exam pairs the two because some categories are so reliably high-risk that re-deciding them each time is both wasteful and dangerous. Deciding them once, ahead of the moment, is what makes the review requirement dependable.
The value of a standing list is that it removes discretion at exactly the point where discretion fails. In the moment, an output looks good and time is short, and that is precisely when a case-by-case call is most likely to waive a review that should have happened. A fixed list is not subject to that pressure.
- Do-not-ship-without-review categories
- A standing list of output types that always require human review before release, decided in advance rather than case by case: (1) final client deliverables, (2) audit-critical or financially material calculations, (3) anything involving regulated, confidential, or highly sensitive data, and (4) public or legal communications where a misstatement carries lasting consequence. The list is fixed and is not waived because an output looks polished.
Final client deliverables
Anything going to a client as a finished deliverable always requires human review before it is released. The client relationship and the professional's name are on the line, and a client-facing error costs trust that is expensive to rebuild. The category is the deliverable's destination, not its apparent quality: a client deliverable is in the category whether it reads roughly or flawlessly, because the review requirement attaches to what it is, not how it looks.
Audit-critical or financially material calculations
Calculations that feed an audit or carry financial materiality always require review. These are the numbers where an error is both consequential and, once relied upon, hard to unwind. The category captures the reason code execution exists as a verification technique: a material figure should be computed and then reviewed by a person, never shipped on a prose estimate or an unchecked computation. Financial materiality is a property of the number's role, so the category applies wherever a figure will drive a material decision.
Regulated, confidential, or highly sensitive data
Any output involving regulated, confidential, or highly sensitive data always requires human review. Here the risk is not only accuracy but handling: obligations of compliance, privacy, and confidentiality govern the content, and those obligations are not removed by using Claude to draft it. The category exists because the consequences of mishandling such data, legal exposure, breach, loss of confidentiality, are severe and often irreversible, so a human check before release is mandatory.
Public or legal communications with lasting consequence
Public statements and legal communications where a misstatement carries lasting consequence always require review. Once public, a statement cannot be quietly retracted, and a legal misstatement can bind or expose the organisation. This category combines high stakes, low reversibility, and often a wide audience, exactly the conditions the risk thresholds flag, but fixes them as a standing rule so the review is never skipped for something that reads well.
What the CCAO-F exam trips candidates on
The first trap is skipping review on a final client deliverable because earlier drafts in the same thread looked accurate. The category attaches to the final deliverable itself, and the accuracy of earlier drafts is not a substitute for reviewing the thing actually being sent. The credited answer reviews the final client deliverable regardless of how the drafting went.
The second trap is treating the list as a soft guideline that can be waived when the output looks unusually polished. The entire purpose of a fixed list is to be immune to the polish that tempts a waiver. The exam rewards treating these categories as always-review, with the output's appearance carrying no weight against the standing requirement.
Worked example
You have iterated a client proposal with Claude across five drafts. The earlier drafts you checked were accurate, and the final version reads beautifully. Under deadline pressure, a colleague suggests sending it without a final review 'since we already checked the earlier drafts.' What is the correct call?
The correct call is that the final client proposal requires human review before it goes out, and neither the accuracy of earlier drafts nor the polish of the final version waives that requirement.
Two traps are in play at once. The first is letting earlier drafts stand in for reviewing the final deliverable. The review requirement attaches to the thing actually being sent, and the final version is not the drafts you checked; iterating changes wording, figures, and emphasis, so a claim that was fine two drafts ago may have shifted. What earlier accuracy tells you is encouraging, not exonerating. The second trap is the deadline-plus-polish pressure to treat the list as a soft guideline. A final client deliverable is a fixed do-not-ship-without-review category, and the whole reason it is fixed in advance is to be immune to exactly this moment, when the output looks great and time is short.
So the output is reviewed as the final client deliverable it is, in full, before release. This is decided by policy, not by how well it reads or how the drafting went. The accountability for the shipped proposal stays with the person sending it under their name, which is why the standing requirement is not the colleague's to waive on a deadline.
Common misreadings to avoid
Misconception
If earlier drafts in the thread were accurate, the final deliverable can skip review.
What's actually true
Misconception
The do-not-ship list is a guideline that can be waived for an unusually polished output.
What's actually true
How this shows up on the exam
Domain 2 questions on this knowledge point present an output in one of the four categories and tempt you to skip review because it reads well or earlier drafts were fine. The dependable answer treats the category as always-review, decided in advance, and refuses to let polish or prior drafts waive the requirement.
This standing list is the fixed complement to the case-by-case four risk thresholds for escalation, and it is applied to concrete situations in applying the escalation thresholds to a scenario. The verification checklist before shipping is the input to these reviews, and the accountability ownership principle is why the review can never be waived away.
After five Claude-assisted drafts of a client proposal, the earlier drafts checked out and the final reads beautifully. Under deadline pressure, should it ship without a final review?
People also ask
Which AI outputs always need human review?
What is a do-not-ship-without-review list?
Why decide review categories in advance?
Watch and learn
Official Anthropic Academy lessons first, then hand-picked walkthroughs. Videos load only when you press play.
No videos curated for this concept yet
We are still curating the best official and community videos for this topic.
Official prep for this domain
Anthropic's own free prep module for this part of the syllabus, on the official prep course. Free with an Anthropic Academy sign-in.
References & primary sources
Master this concept with Archie
Practice it inside an adaptive study session. Archie, your Socratic AI tutor, tracks your mastery with Bayesian Knowledge Tracing and schedules the perfect next review.