# Issues and AI Review

*Audience: everyone who triages or remediates issues raised by Policy Center policies.*

When a policy matches something in your environment, it raises an issue. Relyance AI can then review the issue in context - weighing the evidence behind it - and decide whether it's a considerable risk or a false positive. This page explains what that review shows on an issue, and how it decides which issues you see by default in **Risks & Issues → Issues**.

## The Summary card on an issue

Open an issue that Relyance AI has reviewed and the drawer leads with a **Summary** card, marked with the Relyance AI mark. When the card is collapsed, its header still shows the verdict. Issues that weren't reviewed don't have this card.

The card shows, from top to bottom:

1. **The verdict.** A pill reading **Considerable risk** or **Not a considerable risk**.
2. **Shortfall pills, only when they apply.** A well-supported review shows the verdict pill alone. Extra pills appear only where the review fell short:

   | Pill | Meaning |
   |---|---|
   | **Weak evidence** | The evidence the review had was thin, so its verdict rests on less than usual |
   | **Contradictory evidence** | The evidence pointed in different directions |
   | **Low confidence** | The review wasn't sure of its own call, whatever the evidence |

   Evidence and confidence are separate: the evidence can be clear while the review is still unsure of its call, or the reverse. Where the review recorded a reason for a shortfall, hover the info icon next to the pill to read it.
3. **The key signal.** One line naming the fact the verdict turned on.
4. **How the verdict was reached.** A short explanation in plain language, with links to the places in the code it refers to, where there are any.

### The review trace

Below the summary, click **Details** to expand the trace of how the review was carried out: a bulleted list of what it checked and found, step by step, ending in its decision. The trace is collapsed by default, and only appears when an agent carried out the review. Where the issue comes from data flows, **View in findings** opens the findings behind it.

Use the trace when you want to check the review's reasoning for yourself before acting on it, or to explain to an asset owner why an issue was raised.

### Issues the review confirmed or cleared

The issue's header also carries the Relyance AI mark when the review acted on it:

- **Confirmed by Relyance AI review** - on an open issue the review judged a considerable risk.
- **Cleared by Relyance AI review** - on an issue the review closed as a false positive.

The review only closes an issue when it judged it not a considerable risk **and** its evidence was moderate or strong. A "not a considerable risk" verdict on weak or contradictory evidence leaves the issue open for a person to decide.

## What the Issues list shows by default

By default, the **Issues** list shows only the issues that need your attention:

- Issues the review judged a **considerable risk with high confidence**.
- Issues the review hasn't checked (**Unreviewed**), so nothing is hidden that hasn't been looked at.

Everything else the review produced is still there, under the **AI review** filter:

| AI review filter value | What it includes | Shown by default |
|---|---|---|
| **Considerable risk** | Judged a considerable risk with high confidence | Yes |
| **Unreviewed** | Not reviewed by Relyance AI | Yes |
| **Unconfirmed or low confidence** | Judged a considerable risk with only medium confidence, or reviewed with low confidence whichever way the verdict went | No |
| **Not a considerable risk** | Judged not a considerable risk with high or medium confidence | No |

To see the issues the review wasn't sure about, add **Unconfirmed or low confidence** in the **AI review** filter. Choosing any values in the filter replaces the default, so you see exactly what you picked. Including **Closed** in the status filter also brings back every outcome.

:::info
A considerable-risk verdict now needs high confidence to appear in the default view. Issues judged a considerable risk with only medium confidence moved to **Unconfirmed or low confidence**, so your default list - and the counts that follow it - may be shorter than before. Nothing was closed or deleted by this change.
:::

### Quick filters

Above the list, the quick filters include **Reviewed by AI** and **Cleared by AI**. Hover **Reviewed by AI** to see how many issues the review looked at, how many are still open, and how many it cleared as false positives.

## Working with reviewed issues

- **Trust, then verify.** The verdict is a recommendation. Read the key signal and, for anything high-impact, expand **Details** before acting.
- **Check the hidden queue.** Review **Unconfirmed or low confidence** regularly - these are the issues where a person's judgment matters most.

## Related

- [Admin Configuration Guide](/docs/policy-center/admin-configuration-guide/) - the policies that raise issues, and their remediation steps.
- [Dashboard and Reporting](/docs/policy-center/dashboard-and-reporting/) - posture across obligations and controls.
