Safety Controls | Zyeon UI

Preview in your theme

status='ready' — the everyday panel. Each row is a radiogroup: click a rung or arrow through it and the plain-language line under the scale changes with it. Presets batch-set every row and let go the moment you nudge one; the bar counts the changes and names the ones that LOOSEN filtering. Save fires once no matter how fast you click it, and only the changed categories are in the payload.

Safety filters

Choose how much of each category the model is allowed to return.

Custom — thresholds differ from every preset.

Harassment & bullyingInsults, intimidation and abuse aimed at a specific person or group.

Set by Matthew McGlynn · Mar 12

Threats and sustained abuse are stopped; sharp criticism of ideas still gets through.

Hate speechAttacks on people because of race, religion, gender, disability or origin.

Set by Audrey Block · Feb 28

Anything borderline or worse is stopped — expect some false positives on harmless prompts.

Self-harmEncouragement, methods or instructions for suicide, self-injury or disordered eating.

Every mention of self-harm is routed to support copy, including quotes from novels and lyrics.

Policy floor: “Moderate” or stricter. Required on every workspace; support and crisis resources are returned instead.

Sexual contentExplicit sexual material and sexualised depictions of people.

Severe and moderate content is stopped; mild or ambiguous material still gets through.

Dangerous activitiesWeapons, illicit drugs and instructions likely to get someone hurt.

Set by Dr. Stuart Breitenberg · Mar 03

Only step-by-step weapon and explosive instructions are stopped. Chemistry and firearms history are answered.

Civic integrityClaims about elections, voting procedures and ballot access.

Election claims are returned unchecked — including invented deadlines and polling locations.

no save has fired yet

status='ready' under an org policy. 'Content involving minors' is locked — presets, reset and discard all step over it. 'Dangerous activities' is saved BELOW a floor added later, so the row arrives already staged for the raise and the bar opens with a change nobody made; the rungs under the floor are struck through and skipped by the arrow keys. The 90-character custom classifier wraps inside its row instead of widening the panel.

Safety filters · Acme workspace

Your organisation manages part of this policy. Locked rows and floors are enforced server-side.

Custom — thresholds differ from every preset.

Content involving minorsLockedAny sexualisation or endangerment of a person under 18.

Anything borderline or worse is stopped — expect some false positives on harmless prompts.

Enforced by the provider. This filter cannot be lowered on any plan, by anyone.

Harassment & bullyingInsults, intimidation and abuse aimed at a specific person or group.

Set by Matthew McGlynn · Mar 12

Threats and sustained abuse are stopped; sharp criticism of ideas still gets through.

Hate speechAttacks on people because of race, religion, gender, disability or origin.

Set by Audrey Block · Feb 28

Anything borderline or worse is stopped — expect some false positives on harmless prompts.

Self-harmEncouragement, methods or instructions for suicide, self-injury or disordered eating.

Every mention of self-harm is routed to support copy, including quotes from novels and lyrics.

Policy floor: “Moderate” or stricter. Required on every workspace; support and crisis resources are returned instead.

Sexual contentExplicit sexual material and sexualised depictions of people.

Severe and moderate content is stopped; mild or ambiguous material still gets through.

Dangerous activitiesWeapons, illicit drugs and instructions likely to get someone hurt.

was “Severe”

Severe and moderate content is stopped; mild or ambiguous material still gets through.

Policy floor: “Moderate” or stricter. Raised by your organisation on Mar 14 after a review. The saved value was below it, so the raise is already staged.

Civic integrityClaims about elections, voting procedures and ballot access.

Election claims are returned unchecked — including invented deadlines and polling locations.

Custom classifier — competitor mentions and unverified pricing claims (brand-safety-v3)Your own classifier, trained on 4,200 labelled support replies.

Set by Audrey Block · Mar 14

Only the most severe content is stopped; borderline material still gets through.

1 category changed

no save has fired yet

status='ready' with disabled + presets={[]} + showEffects={false} — the read-only rendering for a member without the permission. Every rung is still readable and the selected one keeps its tab stop, so a keyboard reader can still find out what is set; nothing is clickable and no arrow key moves it, and the panel loses two whole blocks without losing its shape.

Safety filters · read only

You need the Owner role to change these thresholds.

Harassment & bullyingInsults, intimidation and abuse aimed at a specific person or group.

Set by Matthew McGlynn · Mar 12

Hate speechAttacks on people because of race, religion, gender, disability or origin.

Set by Audrey Block · Feb 28

Self-harmEncouragement, methods or instructions for suicide, self-injury or disordered eating.

Policy floor: “Moderate” or stricter. Required on every workspace; support and crisis resources are returned instead.

Sexual contentExplicit sexual material and sexualised depictions of people.

Dangerous activitiesWeapons, illicit drugs and instructions likely to get someone hurt.

Set by Dr. Stuart Breitenberg · Mar 03

Civic integrityClaims about elections, voting procedures and ballot access.

status='loading' — the saved thresholds haven't arrived. Skeleton rows, aria-busy, and the pulse stops under prefers-reduced-motion.

status='empty' — a model that exposes no adjustable thresholds at all. Not an error, and not an editor with nothing in it.

No adjustable filters

This model applies its provider's safety policy and exposes no thresholds you can move.

status='error' — the settings failed to load. No half-open editor: thresholds nobody can trust are worse than none, so the panel offers a retry instead.

Couldn't load your safety settings

The settings service returned 503. request-id: 4c81-2ba9

Loading preview…

This component is part of Pro

The install command, prompt and concept map are included with a Pro license — one purchase unlocks every Pro component, forever.

Unlock with Pro