Hollywood is quietly using AI more than they say.

On October 7, 2025, MPA CEO Charles Rivkin issued a statement demanding OpenAI "take immediate and decisive action" on copyright-infringing Sora 2 outputs. Two months later, on December 11, 2025, Disney (an MPA member) announced a $1 billion equity investment in OpenAI, a three-year deal with one year of exclusivity, and 200+ Disney / Marvel / Pixar / Star Wars characters licensable on Sora for short user-generated videos. The deal never closed. By March 24, 2026 OpenAI had moved to shut down Sora (app closing April 26, API ending September 24, 2026) and Disney had walked away before any money changed hands, saying it respected OpenAI's decision to exit video generation (Hollywood Reporter on MPA statement, Variety on the Sora shutdown and dropped Disney deal, Reuters).

This is the pattern this piece is about. Public restraint and private adoption are operating on parallel tracks at every major Hollywood studio, and the disclosure of AI use is consistently reactive (after press or labor pressure) rather than proactive. For premium AI media buyers trying to scope work for or alongside Hollywood, the gap between the public position and the production reality is the single most important context.

This is an engineering-analyst thesis piece. Every named-production AI use has a primary or near-primary source. Where I could not verify a claim, I flag it.

For the related compliance and legal layer, see AI video and the law in 2026. For the AI media provenance question, see Proving AI media is yours.

What the public position says

The public posture is mostly anti-AI-as-replacement and pro-AI-as-tool, with hard restraint on talent likeness.

• MPA (October 7, 2025): OpenAI must take immediate and decisive action on Sora 2 IP infringement (CNBC).
• CAA (October 8, 2025): "OpenAI/Sora exposes our clients and their intellectual property to significant risk" (Variety).
• WME (October 2025): All clients opted out of Sora 2 by default.
• Japanese CODA (October 27, 2025): Studio Ghibli, Square Enix, Bandai Namco, Aniplex (Sony), Kadokawa, Shueisha jointly demanded OpenAI stop training Sora 2 on their content. Under Japanese copyright law, opt-out is insufficient; prior consent is required (Variety on CODA).
• SAG-AFTRA (September 2025): Publicly condemned the fully AI-generated "actress" Tilly Norwood as "not an actor" but "a character generated by a computer program that was trained on the work of countless professional performers without permission or compensation" (Variety).
• WBD (David Zaslav, May 2024): "The creativity and the kind of empathy and humanity necessary to create exceptional storytelling can only be found in people, not systems."
• James Cameron (March 2025): Avatar 3 will open with a "no generative AI was used in the making of this movie" title card (Variety).

What the production reality is

A verifiable list of named productions that used generative AI in 2024-2025. Every entry has a primary or near-primary source. Most were disclosed reactively after press or fan pressure.

The pattern. Of 13 named productions, disclosure was reactive on 8 (after fan identification or press inquiry), trade-press-only on 3 (Indiana Jones 5, Dune 2, Mufasa), and proactive on 2 (Here, Coca-Cola 2024 after the fact). The default Hollywood mode is to ship AI-augmented work and let press or labor pressure surface the disclosure.

The quiet adoption gap

Beyond named productions, generative AI is in widespread use at the trade-press level across categories that the public never sees on screen.

• Pre-visualisation and storyboarding. AI previs storyboards and animatics are becoming standard in 2025 at major studios per The Wrap's December 2024 reporting. Pixar's Pete Docter confirmed AI for rendering denoising. Netflix launched "INKubator Studio" in October 2025 for AI-generated animated shorts.
• Rotoscoping, cleanup, environment extensions. Foundry CopyCat (used on Dune: Part Two for tattoo removal) and Wonder Studio (Autodesk-acquired) automate rotoscoping. The Wrap reported VFX vendor headcount on tentpoles like House of the Dragon could be reduced by 80 percent or more with these tools.
• ADR and dialogue smoothing. Respeecher is in widespread use (Lucasfilm for young Luke Skywalker in The Mandalorian and Book of Boba Fett; The Brutalist for Hungarian; Cyberpunk 2077 Phantom Liberty for Polish localisation). ElevenLabs has licensed Michael Caine and Matthew McConaughey for voice cloning.
• International dubbing. ElevenLabs Dubbing Studio supports 32 languages. CAMB.AI supports 150+ languages with named Hollywood-adjacent partnerships (IMAX post-production AI dubbing, NBCUniversal partnership). Per market research, 1,100+ feature films globally were AI-dubbed in 2025. Hollywood films released in non-English markets are increasingly AI-dubbed without on-credit disclosure.
• Edit room. Adobe Premiere Pro's Generative Extend (beta October 2024, GA April 2025) is being used by professional editors to extend B-roll, ambient audio, and establishing shots, including in commercial work and increasingly long-form. DaVinci Resolve Neural Engine offers AI-based tracking, noise reduction, and color grading.
• Concept art and production design. Midjourney and Stable Diffusion adoption is widely reported but rarely disclosed on credits.

This layer of AI use is where the math of Hollywood production cost pressure actually lands. Per Hollywood Reporter, US production fell 40 percent from pre-strike levels in 2024. TV episodes dropped from 16,012 in 2022 to 11,069 in 2024. Disney content spend declined from $30B in FY23 to a projected $25B in FY24. Los Angeles County motion picture industry lost roughly 42,000 jobs in the two years between 2022 and 2024 (Hollywood Reporter, Sherwood News).

With content spend compressed by 15-20 percent and production volume down 40 percent, the pressure to use AI for the work that does not require talent likeness consent is mathematical, not ideological. Studios that say publicly they do not use AI while spending $25 billion per year on production are either accepting unsustainable economics or quietly using AI in the categories listed above. Most are quietly using AI.

Studio-vendor partnerships that did not ship as advertised

Lionsgate + Runway (September 18, 2024). Announced as the first Hollywood studio + AI video model deal, with Runway training a custom AI on Lionsgate's catalog. By September 2025 The Wrap reported the deal "hit early snags." Sources told The Wrap and Slashdot: "The Lionsgate catalog is too small to create a model. The Disney catalog is too small to create a model" (The Wrap September 2025). No specific shipped feature or trailer has been publicly tied to the Runway-Lionsgate model.

Disney + OpenAI (December 11, 2025, collapsed March 2026). Announced as a three-year deal: $1B Disney equity, one year of exclusivity, 200+ characters on Sora for short user-generated videos, with talent likenesses and voices explicitly excluded. It was the largest publicly-announced Hollywood-AI commitment of 2025, and it landed against the backdrop of the same MPA member (Disney) signing with OpenAI two months after Rivkin's October statement against Sora 2 IP infringement. It never shipped: on March 24, 2026 OpenAI moved to shut down Sora, and Disney dropped the investment before any money changed hands, saying it respected OpenAI's decision to exit video generation (Variety). Like Lionsgate-Runway, the biggest external studio-vendor AI deal produced no shipped product.

Netflix in-house AI. Eyeline Studios (Netflix-owned VFX house) built the Eternaut building-collapse shot. INKubator Studio announced October 2025 for animated shorts. Netflix is testing Runway tools internally.

The verified pattern: external studio + AI vendor partnerships have not produced flagship shipped product (Lionsgate-Runway stalled, Disney-OpenAI collapsed before launch). Internal in-house AI deployments have shipped (Netflix Eyeline, Disney Mufasa pipeline, Pixar denoising). The IP risk and the cost incentive align with in-house deployment over external partnership.

The "rules for thee, not for me" contradiction

The contradiction is contractually documented.

• TPN v5.3.1 (Motion Picture Association's Trusted Partner Network, launched September 9, 2025) requires studio-vendor security audits with a four-tier Shield system. Most studio contracts in 2025-2026 require TPN Gold for vendors handling content. Trade-press coverage frames AI as one of the emerging threats the system addresses.
• Simultaneously, the same studios deploy AI internally for production (Netflix Eyeline, Disney Mufasa pipeline, Pixar denoising).
• The same MPA whose CEO demanded OpenAI take action on Sora 2 IP infringement in October 2025 has Disney as a member that signed a $1B Sora deal two months later, a deal that then collapsed when OpenAI shut down Sora in March 2026.

This is not hypocrisy in the casual sense; it is the operating reality of an industry under cost pressure. Public restraint protects talent relationships and IP positioning. Private adoption pays for production. The two operate in parallel and the disclosure pattern is consistently reactive.

The talent-side fragmentation

• Tom Hanks (multiple 2024-2025): Publicly mused that audiences may stop caring whether performances are AI; speculated about pitching seven films at age 32 via AI. Defends his use on Here.
• James Cameron (March 2025): Avatar 3 will literally open with a "no generative AI" title card. Cameron is on the Stability AI board. The contradiction is unresolved publicly.
• Bryan Cranston (October 2025): Concern after his deepfake appeared on Sora 2 alongside Michael Jackson likeness. Later joint statement with OpenAI / SAG-AFTRA: "I am grateful to OpenAI for its policy and for improving its guardrails." From concern to cooperation in 30 days, mediated by the union.
• Hayao Miyazaki: 2016 "insult to life itself" statement on AI animation resurfaced during the March-April 2025 OpenAI "Ghibli filter" viral trend. OpenAI added guardrails after the trend.
• Tilly Norwood (September 2025, Particle6): Fully AI-generated "actress" condemned by SAG-AFTRA. Whoopi Goldberg, Emily Blunt, Lukas Gage, Melissa Barrera, Kiersey Clemons among public opponents.

The pattern: talent that signs onto a project with AI baked in (Brody for The Brutalist, Hanks and Wright for Here) defends the use. Talent that wakes up to AI versions of themselves on Sora (Cranston) escalates to the union. The third-rail case is the fully synthetic performer; that aligned union and talent against the technology.

The international fault line

The Japanese position is meaningfully different. CODA's October 27, 2025 letter (Studio Ghibli, Square Enix, Bandai Namco, Aniplex, Kadokawa, Shueisha) is the strongest collective rightsholder action against Sora 2 from any national industry. Under Japanese copyright law per CODA's reading, prior consent is required and the opt-out posture US tech companies have taken is insufficient.

Korea is mixed. K-drama AI dubbing on Amazon Prime Video ("My Man is Cupid," "The Beat of My Heart") sparked Spanish-language backlash in May 2024. Korean production is more conservative on AI than Indian production.

India is aggressive. Abu Dhabi's G42 partnered with Bollywood for AI dubbing, scriptwriting, plot, edit, post. AI dubbing keeps original Shah Rukh Khan or Akshay Kumar voice while translating to Tamil or Telugu. Multilingual releases at scale, recut endings of older titles for additional sales.

China codified the disclosure side. The Cyberspace Administration's 2025 policy requires all AI-generated content, including dubbed voices, to be clearly labelled on-screen and in digital metadata. This is the only major film market with mandatory on-screen AI disclosure as of late 2025.

What this means for premium AI media buyers

Three implications for anyone scoping AI media work in 2026:

One. The public position is not the buying signal. Hollywood studios deploy AI internally far more than they disclose, and the disclosure is reactive. Buyers should not take a studio's public restraint posture as a market signal that AI is not happening. The market signal is the categories of AI use that are economically rational (VFX cleanup, dubbing, pre-vis, edit room) and the failure cases that are publicly contested (talent likeness, fully synthetic performers).

Two. Vendor IP indemnification is the single largest contract lever. Lionsgate-Runway stalled in part because catalog size was too small to train a custom model and in part because of IP risk around the catalog. Adobe Firefly is currently the only major AI media vendor with default IP indemnification on standard paid tiers; that legal posture is becoming the procurement filter. See our AI video and the law piece for the full vendor-indemnification map.

Three. In-house AI is the winning model. External studio-vendor partnerships have not shipped flagship product (Lionsgate-Runway stalled, Disney-OpenAI collapsed). Internal in-house AI deployments have shipped (Netflix Eyeline, Disney Mufasa, Pixar denoising). For AI media buyers scoping a programme, the lesson is that the engineering capability stack should be designed for in-house deployment in the customer's environment, not as a vendor-side dependency. This is the direction we have built our AI content production service around: build the capability inside the customer's environment, with the customer's existing IP and compliance posture preserved.

Keep reading

For the matching legal and compliance map, see AI video and the law in 2026 and proving AI media is yours. For the deep AI video tool comparison and the Sora deprecation playbook, see the AI video tool comparison and Sora is gone. For the broader AI media stack, see AI video generation 2026 and AI dubbing and voice cloning. When you are ready to scope an in-house AI media production stack that maps to your existing IP and compliance posture, our AI content production service covers exactly that, or book a strategy call.