Background
The client is a video generative AI Agent platform: users input text or assets and the platform auto-generates short videos with synthetic human likeness and voice, positioned for content creation and marketing, distributed mainly through app stores. Pre-launch, three compliance obligations triggered simultaneously and interlocked.
Key Challenges
- Deep synthesis filing: Generating synthetic faces/voices requires deep synthesis algorithm filing and prominent user labeling.
- Generative-AI safety assessment: As a public generative-AI service provider, the client must pass a safety assessment covering corpus security, tamper resistance, and refusal mechanisms.
- AIGC labeling: Under GB 45438-2025, generated content needs both explicit (user-perceptible) and implicit (metadata) labels with retention.
- App-store listing: Major stores run a special review for deep-synthesis apps; missing labels mean rejection.
Geason’s Approach
- Filing & assessment in parallel: Mapped the generation chain, clarified the deep synthesis filing entity and the generative-AI safety assessment path, reusing materials across both to shorten the cycle.
- Labeling system design: Built explicit-label templates (badge/watermark placement) and implicit-label rules (file metadata) per GB 45438-2025, with retention and audit mechanisms.
- Listing compliance pack: Delivered a store-review-ready pack: filing evidence, labeling-spec explanation, and a content-safety self-assessment report.
The Non-Obvious Judgment
The industry reduces AIGC labeling to “add a watermark,” but GB 45438-2025 separates explicit and implicit obligations: explicit is user-perceptible; implicit requires metadata writing and retention. Doing only explicit while omitting implicit labeling and retention still fails regulatory sampling and store re-checks. We reframed labeling from a visual action into a fully traceable mechanism mapped to review points.
Key Outcomes
- Deep synthesis filing and generative-AI safety assessment closed, forming a complete evidence chain for the pre-launch three-piece set.
- The AIGC labeling system (explicit + implicit) went live with traceability across generation and distribution.
- The listing compliance pack let the platform pass the special review on first submission.
Client names and commercial data are anonymized. The rules, regulatory procedures and deliverable types described are as they occurred, not simplified.