Blocked Content Policy

1. Purpose

This policy lists the categories of content that Untolds blocks regardless of prompt, persona, setting, or context. It also describes, at a high level, how those blocks are enforced. It complements the prohibited-content list in section 6 of our Terms of Service.

2. Always-blocked categories

The following content is blocked unconditionally. It cannot be unlocked through prompts, settings, payment tier, persona configuration, or any other means:

• Sexual content involving minors— including depictions, descriptions, role-play, suggestions, prompts intended to produce such content, or any attempt to make a persona behave as a minor in a sexual context.
• Non-consensual intimate imagery of real people— including AI-generated "deepfake" intimate imagery of identifiable real persons, and impersonation of real individuals in sexual contexts.
• Bestiality, necrophilia, and content sexualising incest involving minors.
• Terrorism, mass-violence and weapons of mass harm— content that promotes, glorifies, or instructs in terrorism, mass violence, or the production or use of weapons capable of mass harm.
• Self-harm and suicide promotion— content that promotes, glorifies, or instructs in self-harm, suicide, eating disorders, or other serious harm to self.
• Doxxing, threats, hate speech— publishing private contact information of a real person, credible threats against an individual or group, and hate speech against protected groups.
• Real violence / snuff— depictions of real violence against real people, including footage or descriptions of real torture, murder, or fatal accidents.
• Manipulation of vulnerable persons— uses that target a person's age, disability, or socio-economic vulnerability to distort their behaviour, in line with Article 5 of the EU AI Act.

3. How the blocks are enforced

Untolds applies safety controls in layers. Each layer is independent; if one fails the others still apply.

• Input filtering. User messages, prompts, and uploads are scanned at the API boundary before any model is invoked. Matches in the always-blocked categories are refused immediately.
• System-prompt rules.Every persona's system prompt begins with an absolute-rules block that instructs the underlying language model to refuse content in the always-blocked categories regardless of any later instructions.
• Media-pipeline checks. Image, video, and audio generation calls run safety checks on the description, script, or tag set before generation starts. Tags or descriptors associated with blocked categories abort the job.
• Persona-level refusal. Personas are configured with refusal rules that take precedence over user prompts.
• Post-generation moderation. Generated outputs are subject to automated and, where needed, manual review. Outputs matching always-blocked categories are deleted and the originating account is reviewed for further action.

These layers cannot be disabled by users. Attempting to circumvent, "jailbreak", or weaken them is itself a breach of the Terms of Service (section 6) and can lead to immediate account termination.

4. Provenance and synthetic-content marking

All images, videos, and audio produced by the Service are labelled in the user interface as AI-generated and, where technically feasible, marked with machine-readable provenance signals (for example C2PA content credentials or watermarks). Removing or hiding those signals in order to pass synthetic content off as authentic real-world media is prohibited; see section 6 of the Terms of Service.

5. Reporting blocked content that slipped through

No content-safety system is perfect. If you see content that falls into any of the categories in section 2 above and you believe a block failed to catch it, please report it immediately to legal@untolds.chat. For the fastest path to action, see our Content Removal Policy. Suspected child sexual abuse material is treated with the highest priority.

6. Enforcement and consequences

Attempts to produce content in any always-blocked category — successful or not — will result in account suspension or termination, deletion of the offending content, and, where required by law, referral to the competent authorities (including organisations such as NCMEC where applicable). We may also retain a minimal safety record of the event as described in section 9 of the Privacy Policy.

7. Updates

We update this policy when the underlying rules or enforcement layers change in a material way. The effective date above will be updated when that happens.

8. Contact

Questions or reports related to this policy: legal@untolds.chat.