Sora 2 and the end of copyright as we know it (www.technollama.co.uk)

OpenAI quietly launched Sora 2, a high-quality video-generation model that—unlike prior releases—appears to ship with almost no copyright filters, immediately producing viral clips featuring well-known characters (Rick and Morty, SpongeBob, Pikachu, Friday Night/Wednesday Addams mashups). Sam Altman’s pre-launch comment that rights holders would need to opt out now reads like an understatement: the tool’s outputs suggest it was trained on vast amounts of copyrighted video, and users are “stress-testing” the limits by creating obviously infringing content. That surge of examples has put the company squarely in the crosshairs of copyright law and public debate. For the AI/ML community this is consequential. Technically, the release raises questions about dataset provenance, training-phase liability vs. generation-phase liability, and the need for provenance, watermarking, or stronger filters. Legally, OpenAI may be banking on contested defenses—fair use, DMCA safe harbors, tacit licensing deals, settlements, or even political/regulatory protections—but none have been judicially vetted in this context, so risk is high. Practically, we can expect accelerated litigation, new licensing negotiations, and regulatory pressure; at the same time, studios and vendors may adopt or integrate video-generation tools for cost savings. The episode is likely to reshape dataset governance, model deployment safeguards, and the business models around synthetic media.