Back to Seizn

Legal

AI Transparency Disclosure

Review the policies that govern Seizn Author and related services.

Draft pending lawyer review

AI Transparency Disclosure

Last updated: 2026-05-09

Seizn uses large language models (LLMs) and other AI systems to extract canon, detect conflicts, simulate scenes, and generate suggestions. This page explains what is AI-generated, how we deliver disclosure, and what choices you have. We comply with EU AI Act Article 50, California AB 2013, and Korea AI Basic Act §31.

1. AI systems we use

| System | Provider | Use | |---|---|---| | Claude (Opus / Sonnet / Haiku) | Anthropic | Canon extraction, conflict detection, scene simulation, dialog generation | | GPT-4o / GPT-4o-mini | OpenAI (BYOK only) | User-elected fallback or specialty tasks | | Voyage embeddings | Voyage AI | Memory similarity search | | Cohere rerank | Cohere | Result re-ranking on retrieval |

The active provider for each operation is logged and surfaced in the operation receipt (Settings → Activity). For BYOK accounts, the request goes directly to the provider under your account; Seizn passes through but does not retain the content.

2. When you are interacting with AI

We deliver an in-app AI interaction notice the first time you use a feature that involves direct AI interaction (e.g., scene simulation, dialog generation, conflict detection). The notice tells you:

  • That the response is generated by an AI system, not a human
  • Which model is currently routing your request
  • That outputs may be incorrect and should be reviewed before relying on them
  • Where to find the audit log for that operation

You can re-read the notice at Settings → AI → About AI in Seizn at any time. Each delivery is logged in our ai_transparency_events audit table for our own compliance evidence.

3. AI-generated content marking

Content fields produced by AI carry a small "AI" badge in the UI and an ai-generated: true flag in API responses. Examples:

  • Suggested canon entries before you accept them
  • Auto-detected conflicts (until you confirm or dismiss)
  • Simulated scene transcripts
  • AI-suggested mindmap relations

When you export content, the export retains the AI marking unless you explicitly strip it. Stripped exports remain your responsibility under your local labeling laws.

4. Synthetic media (Article 50.4)

If you generate synthetic audio or video using Seizn (e.g., voice samples in scene simulations, when that feature is enabled), the file includes a machine-readable provenance signal (C2PA where supported, or an embedded watermark for legacy formats). The disclosure cannot be removed by us if you re-encode the file outside Seizn; under Article 50.4 you remain responsible for downstream disclosure when redistributing such content.

5. Training data

We do not use your authorial content to train foundation models. We do not provide your authorial content to model providers for training; provider routing operates under zero-retention or contractually capped retention agreements (see Privacy §5).

We may use anonymized + aggregated telemetry (e.g., feature usage frequency) to improve our own retrieval/ranking heuristics. This excludes any prompt content or memory entries.

6. Your rights

  • Opt out of AI-generated suggestions: Settings → AI → Disable suggestions disables candidate generation pipelines while keeping retrieval-only features active.
  • Audit log access: every AI operation is logged with the model, prompt fingerprint, and timestamp. Settings → Activity → Export is always available.
  • Manual review: AI-generated content stays in a candidate state until you accept it. Nothing is committed to your canonical memory without your explicit action.

7. Limitations and known risks

  • Outputs may be factually incorrect, biased, or inconsistent with your prior canon.
  • Long-context simulations may drift from your established voice; review before accepting.
  • Provider outages may degrade output quality (we route to fallback when available; see Status).

8. Effective dates

  • This disclosure: 2026-05-09
  • EU AI Act Article 50 obligations: applicable from 2026-08-02

9. Contact

Status: draft pending lawyer review. Operational binding from launch date.