Guide
Content moderation
Uncensored models are subject to our content moderation policies. These policies cover child sexual exploitation and abuse, mass-casualty violence and weapons, and fraud and extortion.
What we moderate
We moderate requests that seek the following content or assistance:
- Child sexual exploitation and abuse. Sexual or sexualized content involving anyone under 18, including fictional depictions; grooming or gaining access to a child for sexual purposes; and creating, obtaining or sharing sexual material depicting minors.
- Mass-casualty violence and weapons. Assistance to carry out attacks intended to cause mass casualties, or to make, obtain or use chemical, biological, radiological or nuclear weapons or explosives capable of mass casualties. This includes actionable technical instructions and sourcing assistance.
- Fraud and extortion. Assistance to steal money, credentials or property through deception; run scams, identity theft, money laundering or counterfeiting operations; or blackmail and coerce someone through threats.
Why do uncensored models still need content moderation?
We want people to use these models as freely as possible. Our restrictions target a small set of abusive uses, including child exploitation, mass-casualty attacks, fraud and extortion.
Broad refusals can block legitimate work simply because it involves sensitive subjects. A narrower boundary preserves more useful applications while targeting the abuses we want to prevent. Removing broad refusals and enforcing that boundary serve the same goal: greater freedom for legitimate use.
Longer term, we'd like to develop our own abliterated models that retain refusals for these specific abuses while allowing other legitimate uses.
How do you moderate uncensored models with ZDR?
We host our own ZDR classifiers alongside the LLMs to moderate content without storing prompts or responses. Subscribers can enable ZDR for uncensored models. A flag is an automated assessment and can be mistaken.
Receiving a response does not mean your request complies with our policies or that you cannot be suspended later. Repeated flags can suspend your organization's access to uncensored models.
See Data controls for uncensored model retention and ZDR settings.
How long suspensions last
Each new suspension escalates: 1 hour, 24 hours, 7 days, 30 days, then permanent. The same schedule applies separately to an organization and to each attributed reseller end customer. Settings › Organization shows your organization's end time. The dashboard banner and an email to your organization's admins include the UTC unban time, or say when a suspension is permanent. Other available models remain accessible.
A suspension does not grow longer while it is active. After it ends, only new requests can trigger the next suspension. Expiry or an early release does not reset the schedule. You can appeal any suspension, including a permanent one.
Requesting a review
If you believe your access was suspended in error, open Settings › Organization to submit an appeal and follow replies from the review team. You can also email founders@experientiallabs.ai with your organization ID, any relevant request IDs, and a short explanation. See the KYC guide for requests to restore access early. You can appeal a temporary suspension before its end time. An approved appeal can lift the restriction early.