Sora 2 Arrives: OpenAI Launches Its Own 'AI TikTok' With Viral Cameos

OpenAI released Sora 2 on September 30, 2025 — markedly more physically accurate and the first with synchronized audio — alongside an invite-only social app: a TikTok-style AI video feed whose signature Cameo feature drops you and your friends into any AI video.

The app shot to the top of the U.S. App Store, but controversy followed: copyrighted-character abuse and celebrity deepfakes forced OpenAI to repeatedly tighten generation policies.

Cameo: Turning 'Put Me in the Video' Into a Social Hook

The app's breakout wasn't model quality — it was Cameo: with consent, your likeness and your friends' could be dropped into any AI video. That turned generative video from a creation tool into social currency, and friends casting each other built a natural distribution chain. Invite-only scarcity plus the cameo mechanic sent it to #1 at launch and made 'AI TikTok' the hottest product narrative of the season.

Why the Controversy Wouldn't Die Down

Problems arrived as fast as the hype: copyrighted characters generated at scale, deceased celebrities re-animated for entertainment, strong pushback from Hollywood and Japanese IP holders — OpenAI retreated from default-allowed to rights-holder opt-in. The tug-of-war belongs to the same storyline as the NYT lawsuit and the Anthropic settlement: the rights boundaries of generative content remain unsettled (see our policy coverage).

The Ending, Seven Months Later

Per OpenAI's help center, the Sora web and app experiences were discontinued on April 26, 2026, with capabilities folded into other products — the splashy AI-social experiment lasted roughly seven months. The model itself did not retire: Sora 2 capabilities continued via the API and ChatGPT. What exited was the standalone-social-platform form factor.

Our Take

Sora's seven months make a complete stress-test sample: model capability can detonate growth, but a content platform's moat lives in rights infrastructure, creator economics and trust mechanisms — none of which model iteration speed can fix. The lesson for the next builder: the battleground for AI content platforms is not generation quality but the design of rights clearance and revenue sharing.