> For the complete documentation index, see [llms.txt](https://docs.cortex.aic.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.cortex.aic.io/workflow-guides/products/shield.md).

# Shield

**Open this area:** In the CORTEX menu, select **Assurance & governance → Shield sanitisation**.

Remove explicitly selected terms from one retained Product revision, review the result and preserve unchanged handling and protected original lineage.

**Who uses this:** Author, Analyst, Reviewer.\
**Open in CORTEX:** `/shield`.\
**In-app guide:** `/help?topic=shield`.

**Expected result:** A separate immutable internal derivative is saved without changing the original or granting release authority.

## Before you start

* The page shows whether Shield, Product, Memory and Analyse are enabled and whether your current grant includes sanitisation and source access.
* Choose an original Product and exact retained revision; chained derivatives are not supported. Restricted originals need a current approved processing decision for their original datasets.
* Know the specific literal terms to remove and why, without assuming this will detect other sensitive meaning.

## What to do

{% stepper %}
{% step %}

## Inspect the original

The first step needs one visible original and exact retained revision. Use Find original products, inspect its provenance and fixed handling, and check the processing prerequisite before continuing.

The service rechecks source access and approval on creation.
{% endstep %}

{% step %}

## Define literal rules

The second step marks terms and reason required and shows remaining errors as you type.

Enter 1–20 distinct protected literal terms, one per line, each 3–128 characters and present in the selected original text, citation, source note or processing assessment.

Give a 10–500 character reason.
{% endstep %}

{% step %}

## Review the preview

The final step needs Preview literal removal and explicit confirmation.

Inspect title, headings, body and retained assurance notices. Original citations, source notes and optional analysis content are omitted from the derivative's output, while protected lineage remains enforced internally.
{% endstep %}

{% step %}

## Create the separate record

Confirm the literal-only result and unchanged handling, then select Create immutable sanitised draft.

Check the saved Product rather than treating the unsaved preview as a retained derivative.
{% endstep %}

{% step %}

## Check actual output and release

Inspect the derivative through normal Product export. Literal exclusions are checked against generated text and metadata; resolve any refusal without dropping mandatory warnings.

Use independent controlled release if dissemination is required.
{% endstep %}
{% endstepper %}

## Check the result

* Is the selected original content revision correct?
* Have I inspected all retained text and mandatory notices?
* Could protected meaning remain through paraphrase or context?
* Is the unchanged handling suitable for the intended audience?
* Have I kept sanitisation separate from release approval?

## If you get stuck

<details>

<summary>Why cannot I edit the sanitised draft?</summary>

It is deliberately immutable. Return to the appropriate original revision and create a new separately reviewed derivative with the correct rules; the old derivative and original remain retained.

</details>

<details>

<summary>Why is a term or rendered export refused?</summary>

Check term length, uniqueness and whether it occurs in the selected source content. A required marking or notice may collide with an exclusion; do not remove mandatory controls to force output. Reassess the intended sanitisation with the responsible authority.

</details>

<details>

<summary>Can this make a lower-marked report for wider release?</summary>

No. Shield preserves the complete handling and source-access obligations. Seek an approved organisational disclosure/down-marking process; no such authority is supplied by this control.

</details>

## Responsibility and limits

{% hint style="warning" %}
A human evaluates disclosure risk beyond literal matches. Shield does not recognise sensitive meaning, images or every possible identifier.
{% endhint %}

{% hint style="warning" %}
Removing words does not lower handling or make an otherwise ineligible recipient eligible; original access obligations continue.
{% endhint %}

**Reference and scope:** Implementation guidance supporting controlled output in JDP 2-00 (4th ed.), 3.49–3.59 and 4.33–4.45. Literal sanitisation does not constitute declassification or disclosure approval.

**Related workflows:** [Author an intelligence Product and inspect its exports](broken://pages/262a7998f4bd8e3a741063bf243590ce8bf23fc9); [Review, approve and publish an exact internal release](broken://pages/470978e3be0758fbc43448d1d7b8fc918bd9675f); [Preserve an exact revision and manage holds](file:///7397340/assurance/governance.md).