Green/Yellow/Red Data Sensitivity Tiers for the CCAO-F Exam

Classify before anything is uploaded

Before any data enters any Claude feature, you should already know its sensitivity. The CCAO-F exam treats data classification as the habit that prevents leakage, and it is deliberately a fast, three-tier scheme so that it can be applied in seconds on every upload rather than reserved for the occasional obviously-sensitive file. Green, yellow, red: safe to use, review first, keep out. The whole point is that the judgment happens before the data reaches a feature, when you can still stop it.

Three tiers are enough because most decisions do not need more resolution than that. What matters is that the practitioner runs the check every time, not that the taxonomy is elaborate. Getting the boundaries between the tiers right - especially the ones people misjudge - is the substance of this knowledge point.

Green: safe to use

Green data is information whose exposure carries no meaningful risk because it is already public or has already been cleared. Published material - anything the organization has released to the world - is green. Anonymized or aggregated data is green because the sensitive specifics have been stripped or rolled up so no individual is exposed. And internal content already cleared for wide sharing is green, because someone has already made the judgment that it is safe to circulate.

The unifying idea is that green data has either never been sensitive or has already passed the check. It needs no special control and can go into a feature with only the ordinary care you give any work. The trap on the green tier is the reverse of caution: assuming something is green when it has not actually been cleared.

Yellow: review first

Yellow data is the middle tier, and it is the one that catches the most everyday material. Three kinds of content land here. Internal-only material - documents meant to stay inside the company - is yellow because it was never intended to leave. Anything with names or contact details is yellow, because personal identifiers raise privacy stakes even when the surrounding content seems mundane. And unannounced deal or product content - drafts tied to something not yet public - is yellow because premature exposure carries real business and legal risk.

Yellow does not mean forbidden; it means review first. The content needs a deliberate check against policy before it goes anywhere near a feature, and often a matching control (such as keeping it out of persistent Memory) once cleared. The common error here is treating "not public" as equivalent to "fine," which collapses yellow into green and skips the review the tier exists to trigger.

Red: keep out unless an approved path exists

Red data is the tier where the default is out. Three categories define it. Regulated data - health, financial, government information governed by law - is red because mishandling it can breach legal obligations. Credentials and secrets are red because their exposure directly enables compromise. And data under a third-party confidentiality obligation is red because the organization has promised someone else to protect it.

Red data does not go into a feature on individual judgment. It stays out unless an approved, confirmed path exists for exactly that kind of data. The sharpest trap on this tier is treating credentials or secrets as merely yellow because they are not personally identifiable - a password is not PII, but it is unquestionably red, because the harm from exposing it is severe regardless of whether it names a person.

What the exam trips candidates on

The first trap is assuming any internal document is automatically fine to upload because it is not public. Non-public is not the same as safe. Internal-only material is yellow and needs review, and internal content can even be red if it contains regulated data or secrets. The "it's just internal" reflex is exactly what the tiers are designed to interrupt.

The second trap is treating credentials or secrets as merely yellow because they are not personally identifiable information. Sensitivity is not only about PII. A secret's risk comes from what it unlocks, not from whether it names an individual, so it sits firmly in red no matter how impersonal it looks.

Common misreadings to avoid

How this shows up on the exam

Domain 6 questions on this knowledge point give you a piece of data and ask which tier it belongs to or how to handle it. The reliable approach is to match the item to the tier's defining categories - published/anonymized/cleared for green, internal/personal/unannounced for yellow, regulated/secret/third-party-confidential for red - and to resist the two misfiling reflexes about internal documents and non-PII secrets.

These tiers are the foundation for the rest of Task 6.2. When you are unsure which tier applies, the default-to-more-sensitive rule tells you which way to round; redaction and anonymization can sometimes move data down a tier; and the feature-level data controls are how you act on a tier once it is assigned.