September 30, 2025 — OpenAI today formally launched its short‑video creation and sharing platform Sora 2. The app allows users to generate short videos from text and uploaded personal assets (cameos) and share them on social networks. OpenAI calls it an important step toward "AI video generation entering the mass‑application stage," but media and industry immediately focused on a sensitive point: it allows videos to be "spun from" existing copyrighted content, which may appear in the platform feed even if rights holders have not explicitly authorized use.
1. Product and Feature Highlights
- Spun from copyrighted content
Reports indicate Sora's recommendation feed may include content "spun from copyrighted films or materials" unless rights holders actively "opt out". In other words, OpenAI provides an opt‑out for rights holders who do not want their works used for model‑derived generation; but by default the model may use public or flagship materials as a "training trigger."
- Cameo and consent mechanism
To reduce abuse risks, Sora introduces a so‑called Cameo mechanism: if a user wants to include a person's likeness (facial/identity features) in a generated video, that person must first upload their "consent material" and pass live verification (for example, turning the head or speaking a random sequence of numbers). During the draft stage, the user whose likeness is involved can also view drafts that include their likeness to confirm consent.
- Time limits and clip length
Sora currently limits generated videos to a maximum of 10 seconds. This restriction may stem from model stability concerns, abuse prevention, or copyright control considerations.
- Rights‑holder "opt‑out first" policy
Similar to image‑generation policies, OpenAI sets Sora's feed policy so that copyrighted works are available for model‑derived use by default unless rights holders explicitly opt out. Reports note that one major studio—Disney—has already decided to opt out of having its content used for Sora derivations.
- Protection of public‑figure likenesses
OpenAI says that for public figures and individuals who have not uploaded a cameo, Sora will disable generation of their likeness unless they upload verified material. In other words, synthesized videos should not contain a public figure's likeness without permission.
- Live‑check mechanism
- To prevent unauthorized use of photos or static material to impersonate someone, Sora requires users to pass a live‑check (e.g., head turns plus speaking a random number) to confirm identity authenticity.
2. Copyright Disputes and Compliance Risks
1. Default use vs. opt‑out rights
OpenAI chose to make copyrighted works available for model "derivation" by default unless the rights holder proactively opts out. This "use‑first, withdraw‑later" approach may be seen as encroaching on rights holders' interests. Many film studios and music rights holders may not want their works to become part of a model's material pool. Aside from Disney's stated opt‑out, it remains to be seen whether other major content providers will follow.
2. AI training and the "fair use" claim
In related reports, OpenAI has lobbied the U.S. government to recognize training AI on copyrighted material as falling under "fair use" to provide legal protection for future model training and generation. That claim is contentious among legal scholars, rights organizations and the creative industries: even if training is deemed fair use, whether generated outputs are exempt or how copyright in those outputs should be allocated remains unresolved.
3. Defining derivative works vs. authorship
When a model generates a piece based on a film's style or visual cues plus prompt text, is it a wholly new work or a derivative work? The perennial adaptation/derivative question in literature, film and music is being rewritten in the AI era. Who owns the copyright in model outputs? Must creators pay the original rights holders? These issues are likely to become focal points for industry disputes in the near term.
3. Impact on the Creative Industries
1. Lowering production barriers and reshaping short‑video ecosystems
If widely adopted, Sora could dramatically compress production costs for videos, animation, ads and trailers. For small creators, MCNs and brands, a short creative clip can be generated in minutes for rapid validation and iteration. Traditional production roles—photography crews, actors, set construction—may be squeezed in low‑ to mid‑budget projects.
2. Restructuring the rights and stock‑asset market
Values of traditional asset libraries (stock footage, music, sound effects, image banks) could be depressed because many needs may be met by model generation. Operators of asset libraries will need to evolve into "model‑friendly asset providers," partnering with generation platforms to offer licensed, trainable assets and identity‑verification services as value‑added offerings.
3. Shifts in competition between platforms and creators
Sora itself has social and content‑platform attributes and may compete directly with short‑video platforms like TikTok, YouTube Shorts and Instagram Reels. Morgan Stanley analyst Brian Nowak noted in a research report that the Sora app could capture user attention and compete with Meta, Google and TikTok's content ecosystems.
4. Deepfake Risks: Scaling and Mainstreaming
Sora's launch means realistic synthetic audio‑visual content may move from isolated research efforts to mass distribution on consumer devices. Potential risks include:
- Fabricated evidence / influence operations: Synthesized videos of political figures in staged scenarios could mislead public opinion, especially around elections or public events.
- Social scams / identity fraud: Impersonation via generated videos could be used for fraud, extortion or defamation.
- Privacy and personality‑rights violations: Generating videos of individuals without authorization, particularly involving sensitive content, can violate privacy and reputation.
- Abuse involving minors: Generation tools can be misused to create inappropriate content involving minors, an area of high legal sensitivity.
Although OpenAI has built defenses such as cameo verification, upload permissions and live checks, these mechanisms can still be bypassed or attacked. Technical countermeasures (e.g., watermark removal, realism optimization) also form part of an adversarial cycle.
5. Regulatory and Legal Challenges: Boundaries Not Yet Set
1. Copyright ownership and liability for AI‑generated content
Under current legal frameworks it is difficult to determine who owns model‑generated videos—the prompt author, the model provider, or a combined author. Liability for infringement or defamation arising from such content is likewise unclear.
2. Legality review of training materials
Sora's default allowance for using copyrighted material as a basis for derivation implies that OpenAI and training partners must rigorously review the legality of their training corpora to mitigate potential copyright litigation risk.
3. Legal watermarking, provenance systems and accountable design
For regulatory and judicial evidence purposes, mandatory "robust watermarking / provenance" systems may be required. Generation platforms could be obliged to embed tamper‑resistant watermarks or identifiers in every video to enable future accountability.
4. Youth users and content moderation policies
Regulators may demand stricter review standards, age verification and child‑protection measures for generated content that involves minors or sensitive scenarios.
6. Balancing Paths: OpenAI's Measures and Industry Responses
OpenAI's defensive design
- Opt‑out mechanism: Rights holders can proactively opt out of having their works used by the model.
- Verification and upload policy (Cameo + Live Check): Use of likeness requires the person to upload verified material and pass live checks.
- Draft visibility: Individuals can view drafts that include their likeness and explicitly consent.
- System cards and use disclaimers: OpenAI has published transparent system cards that prohibit certain uses of synthesized public‑figure videos.
- Monitoring and abuse response: The platform includes abuse monitoring and complaint mechanisms.
Industry and policy roles
- Rights holders: Should proactively monitor and choose early opt‑out or negotiate terms to protect their works.
- Content platforms: Must apply safety checks and moderation when using AI‑generated videos, not allow indiscriminate publication.
- Regulators / legislators: Need to accelerate legal definitions for AI‑generated video, especially around deepfakes, personality rights and copyright liability.
- Creators / tool providers: Can adopt "model‑friendly licensing + creator incentives" to collaborate with platforms and safeguard the creative ecosystem.
7. Outlook: The Trade‑off Between Innovation Dividends and Governance
The launch of Sora 2 marks AI audio‑visual generation's entry into the mainstream. It could reshape short‑video production, accelerate creative‑industry innovation and lower entry barriers. At the same time, it brings deepfakes, copyright ambiguity, misuse risks and platform‑governance challenges to the fore. OpenAI's design offers partial templates—compliance verification, opt‑out, transparent policies and abuse monitoring. Over the coming months, as Sora 2 scales, developers experiment and litigation emerges, we will gradually see the technology's real‑world boundaries, derivative responsibilities and governance mechanisms take shape.