OpenAI just proved that building on closed AI infrastructure is a bet with an expiration date — and Hollywood's most ambitious filmmakers are learning this the hard way.
The Summary
- Nik Kleverov, director of *Critterz*, was building one of the first AI-generated feature films using OpenAI's Sora when OpenAI abruptly shut down the tool mid-production
- The film was initially slated for Cannes 2026 but got derailed by technical issues (read: vendor lock-in risk)
- Kleverov still believes 2026 is "year zero" for mass adoption of AI in film, arguing human storytelling talent matters more than ever
- The case study exposes a critical tension: AI tools promise creative independence, but closed platforms deliver dependence
The Signal
*Critterz* represents something more significant than a single derailed film project. It's the first high-profile case study of what happens when creative professionals build entire productions on proprietary AI infrastructure controlled by a single vendor. Kleverov and his production company Native Foreign partnered with OpenAI after creating a short film using DALL-E in 2022, then moved to Sora for the feature. The relationship looked like the future of filmmaking: a major AI lab backing creative talent with cutting-edge tools. Until OpenAI killed the platform.
The technical details matter here. Sora wasn't deprecated gradually or replaced with a better version. It was shut down while active projects depended on it. This isn't about a tool getting better. This is about the fundamental instability of building on closed AI platforms where the roadmap serves the platform owner, not the creator.
"The core technology we were building the film with went away in the middle of making the film."
What makes this particularly revealing is the timing and context. Before the technical shutdown, French groups were already threatening to boycott *Critterz* at Cannes. The cultural resistance to AI-generated films is real and organized. But the bigger threat turned out to be infrastructural, not ideological. You can negotiate with film festival programmers. You can't negotiate with a platform that decides your production tools no longer align with their product strategy.
Key inflection points:
- 2022: DALL-E enables experimental shorts
- 2023-2024: Sora promises feature-length production capability
- 2025-2026: OpenAI shuts down Sora, stranding productions mid-workflow
- 2026: Kleverov still calls this "year zero" for AI in film despite the setback
Kleverov's framing of 2026 as year zero is either delusional or strategic. He argues human talent is more valuable than ever when AI tools proliferate, which is the standard Hollywood cope. But the subtext is more interesting. If you believe AI video generation is inevitable and mass adoption is coming, then early platform casualties like *Critterz* are just the price of being first. The question is whether the next wave of AI filmmakers learns the right lesson.
The right lesson isn't "don't use AI tools." It's "don't build critical infrastructure on platforms you don't control." Open source models, locally run inference, and composable toolchains are the only way to avoid getting rug-pulled mid-production. The film industry has spent a century building on stable technology platforms, cameras and editing software that don't disappear because a VP in San Francisco changed the roadmap. AI-native filmmaking needs the same stability, and closed platforms cannot provide it.
The Implication
If you're building anything production-critical on closed AI platforms, *Critterz* is your warning shot. OpenAI didn't kill Sora because the technology failed. They killed it because it didn't fit their evolving strategy. That's the risk profile of closed infrastructure: your creative timeline is hostage to someone else's pivot.
The path forward splits in two directions. Hollywood can wait for stable, proven tools backed by companies committed to long-term creative partnerships. Or early adopters can embrace open source models and accept the overhead of running their own infrastructure. The middle ground, betting your production schedule on a startup's beta product, just got a lot riskier. Kleverov says 2026 is year zero. He's right, but not for the reasons he thinks. This is year zero for understanding that AI-native creative work requires AI-native ownership of your tools.