Skip to main content

Acceptable Use Policy (draft outline)

Updated October 10, 2026

Draft — pending legal review; not yet in effect

This document is a working draft prepared for counsel. It may change before launch and is not yet in effect.

Draft for legal review — not in effect.

This outline separates rules that the product already enforces as gates from rules that need policy text and human enforcement. Text in [COUNSEL: …] needs legal drafting. The everyday version for customers is Prohibited uses.

1. Purpose and scope

  • Applies to every user, workspace, API key, MCP client, automation and agency client account. [COUNSEL: incorporation into the Terms].
  • Principle used across the product: "transparent fiction, not deception" — AI personas are allowed when they are disclosed as AI and do not deceive.

2. Rules the product enforces today (hard gates)

These cannot be overridden by any user in the app. AI quality review marks them as hard gates and approval or publishing is refused with a compliance block.

Rule How it is enforced
AI disclosure for synthetic personas, AI voices and digital twins A disclosure marker must be present in the caption, on screen or spoken; otherwise the content fails review and cannot be approved
Endorsement disclosure (sponsored, gifted, affiliate, own product) Missing endorsement disclosure is a hard gate
Prohibited or unsubstantiated regulated claims; individualized health or financial advice Claim checks fail; [CLAIM: …] tags must match approved brand or script claims with evidence
Unlicensed commercial music or footage Hard gate in review
Real person's likeness or voice without recorded consent Persona consent gate; voice cloning stays disabled until a consent is recorded
Minors' likeness Any consent whose subject is a minor blocks every use; personas with an age range under 18 are flagged
Public-figure look-alikes Screening match flags the persona and blocks the workflow
Rule-pack BLOCK findings (jurisdiction or platform rules) Every BLOCK finding is a hard gate
Consent revoked The persona's assets are marked revoked and open approvals are invalidated, so affected content needs a new review before it can publish (the publish worker re-checks approval before posting)

3. Rules that need policy text and human enforcement

  • [COUNSEL: prohibited content categories — illegal content, sexual content involving minors (absolute), non-consensual intimate imagery, hate, harassment, violent extremism, self-harm promotion, weapons, regulated goods].
  • [COUNSEL: deception — impersonation of real people or organizations, fake reviews or testimonials, astroturfing, coordinated inauthentic behaviour, election and political content rules].
  • Product fact: the product blocks first-person testimonials for experiences a fictional persona did not have and invented credentials or results only through review checks and rule packs; [COUNSEL: state this as a rule].
  • [COUNSEL: high-risk verticals (health, finance, legal, alcohol, politics, child-directed) — allowed only as education, with specialist review; the product applies stricter automation policy and requires specialist review in these verticals].
  • [COUNSEL: intellectual property — only upload content the customer owns or has licensed].
  • [COUNSEL: platform rules — customers must follow each social network's terms; no scraping or bypassing network access controls (the product never uses credential scraping)].
  • [COUNSEL: spam and lead capture — consent required; the product records consent per purpose with double opt-in and withdrawal].

4. Security and technical misuse

  • [COUNSEL: no probing, load testing or vulnerability scanning without written permission; no circumventing rate limits, credit limits or spending caps; no sharing API keys; no reselling without an agency agreement].
  • Product facts: rate limits apply per user and per IP; API keys are scoped and can be revoked; webhooks are signed; website reading respects robots.txt and refuses private network addresses.

5. Automation and AI agents

  • Automations, Autopilot, Copilot and MCP clients are bound by the same rules. The customer sets an automation policy, budgets, required approvals and kill switches. [COUNSEL: customer remains responsible for automated actions].

6. Enforcement

  • Product facts: workspace owners and admins can suspend members; platform operators can disable models, providers and feature flags per workspace, and every admin action is audited. A moderation-flag table exists, but no workspace-suspension or moderation console for AUP cases exists yet [Eng: build before launch, or COUNSEL: manual process]. [COUNSEL: notice, appeal, suspension, termination, reporting to authorities].
  • [COUNSEL: abuse reporting address — none is wired into the app yet (launch checklist 5.2)].

More in Legal