Is it illegal to use AI to make videos?

The Legal Gray Zone: Why AI Video Creators Are Confused

Here's the short answer: making AI videos is not illegal in the United States — but "not illegal" and "legally safe" are two very different things right now. The law around AI-generated video is unsettled, actively contested in court, and shifting fast enough that what's defensible today may be risky next year.

The scale of the uncertainty is striking. According to the Copyright Alliance's 2025 lawsuit tracker, more than 70 AI copyright infringement lawsuits are now pending against AI companies — more than double the roughly 30 cases on file at the end of 2024. Litigation now spans the entire AI value chain, from dataset providers and model developers to everyday users of AI outputs, as tracked by BakerHostetler's case tracker.

What makes 2025 a turning point for video specifically? Hollywood entered the fight:

• Disney and Universal sued Midjourney in June 2025, with Warner Bros. filing a similar complaint in September.
• Disney, Universal, and Warner Bros. jointly sued China-based Minimax over its Hailuo AI video generator, alleging it can output protected characters like Star Wars, The Simpsons, and Shrek.
• Video creators sued ByteDance/TikTok in December 2025 for scraping millions of YouTube videos to train its "MagicVideo" model — under DMCA anticircumvention provisions rather than direct infringement.

These are the first-ever film studio lawsuits against AI video generators, and they signal that AI video tools — not just text and image models — are now directly in the crosshairs.

OpenAI's Sora app, launched September 30, 2025, added fuel to the fire. A Harvard JSEL analysis notes that OpenAI's policy requires copyright holders to opt out through individual requests, with no blanket protections — meaning smaller creators "may be left in the dark." The author predicts Sora "will likely lead to a flurry of litigation."

For business owners and creators, this gray zone creates real operational questions. Which video tools carry more legal exposure? How much human editing makes a clip protectable? And if you're building AI workflows with agents and skills, how do you know which third-party components are safe to use — a problem SkillzBench is built to address with security-scanned, license-checked agent skills. The confusion isn't a reason to avoid AI video; it's a reason to understand exactly where the legal lines currently sit.

Who Owns Your AI Video? The Human Authorship Rule

You can generate a video with AI in seconds — but under U.S. law, that video may belong to no one at all. That's the counterintuitive reality creators face after the D.C. Circuit's March 2025 ruling in Thaler v. Perlmutter, which affirmed that the Copyright Act "requires all eligible work to be authored in the first instance by a human being" (per the Congressional Research Service). Purely AI-generated video falls into the public domain the moment it's created.

The court's reasoning is grounded in the statute itself. The D.C. Circuit pointed to the provision setting copyright duration at 70 years after the author's death — a timeline that only makes sense if the author is human (CRS analysis notes). The Copyright Office has refused to register purely AI-generated works, and a federal district court has upheld that position (according to the Copyright Alliance).

Many creators assume that writing a detailed prompt equals authorship. The Copyright Office rejected that view in its January 2025 report, concluding that "prompts alone do not provide sufficient human control to make users of an AI system the authors of the output." The Office analogizes an AI user to "a client who hires an artist" who gives only general directions — you commissioned the work, but you didn't create it (per CRS).

The Zarya of the Dawn precedent made this concrete. The Office granted, then partially canceled, copyright for a Midjourney-assisted graphic novel — protecting the human-written text and arrangement while excluding the AI-generated images, because prompts produce "unpredictable outputs" and the edits were "too minor and imperceptible" to qualify (as Built In reports).

The good news: hybrid works are not lost causes. The Copyright Office has registered hundreds of works incorporating AI-generated material, with protection covering the human author's contribution (per CRS). Human authorship can be established through "minimally creative arrangement, selection, and coordination" of AI-generated parts (per the Copyright Alliance).

To build a protectable AI video:

• Substantially edit, select, and arrange the AI-generated footage rather than accepting raw output
• Contribute original elements — script, narration, music, sequencing — that reflect your creative control
• When registering, identify and disclaim the AI-generated portions; failure to do so led to registration denial in *Théâtre D'opéra Spatial*

As Vanderbilt law professor Daniel Gervais puts it, copyright "will only focus on the human part" — and it "really needs to be an authorial kind of contribution" (per Built In). For teams building AI video workflows, the lesson is clear: meaningful human creativity is what converts a public-domain output into an asset you actually own.

Where You Can Actually Get Sued: Infringement Risks in AI Video

In 2025, a court ruled that training AI on copyrighted books was fair use — and the same AI company still paid the largest copyright settlement in U.S. history. That paradox is the key to understanding where your AI video project can actually get you sued.

The two rulings in question, Bartz v. Anthropic and Kadrey v. Meta, held in June 2025 that training generative AI on lawfully accessed books was "quintessentially transformative." Yet Anthropic still settled for $1.5 billion — roughly $3,000 per work — because it had downloaded roughly 482,460 pirated books to build its training library. The lesson: how a model obtained its training data is a separate legal question from whether training itself is fair use. If your video tool was built on scraped or pirated material, you're using a product with proven liability attached.

Outputs are an even bigger risk. Film studios filed their first-ever lawsuits against AI generators in 2025: Disney and Universal sued Midjourney in June, Warner Bros. followed in September, and all three sued China-based Minimax, alleging its Hailuo AI can generate videos featuring protected characters like Star Wars, Shrek, and The Simpsons. Under the Supreme Court's Warhol v. Goldsmith framework, a recreation serving the same commercial purpose as the original — entertainment — weighs heavily against fair use. As one Harvard analysis put it, even if training is fair use, the output may not be.

So what actually puts you on the wrong side of a courtroom?

• Generating recognizable copyrighted characters, like Disney, Universal, and Warner Bros. allege against Midjourney and Minimax
• Producing videos that compete with original works in their existing markets — the Copyright Office's May 2025 report says fair use doesn't apply there
• Relying on tools with opt-out-only policies; OpenAI's Sora requires copyright holders to request protection one by one, with no blanket options
• Assuming fair use covers you because it covered a model trainer — the Kadrey court explicitly called its ruling narrow

Your tool choice shapes your exposure. The market is splitting into opt-out models like Sora and licensed models like the ones Suno plans for 2026, plus royalty-paying startups such as Bria, which trains only on "responsibly sourced" data. With over 70 AI copyright lawsuits now pending — more than double the ~30 at the end of 2024 — picking a tool with a clear licensing posture is becoming part of due diligence, not just a feature comparison.

The same scrutiny applies to the building blocks around your AI workflow. When you're stacking agent skills and tools onto your video pipeline, unvetted components carry their own license and security risks — research has found roughly 26% of published agent skills carry at least one vulnerability. That's why SkillzBench is building license-checking, security scanning, and benchmarking into its free skills directory, so you can see what you're attaching before you attach it.

The law here is a moving target — key fair-use questions remain undecided, and courts aren't bound by Copyright Office interpretations. Treat today's rules as provisional, and build accordingly.

A Practical Compliance Playbook for AI Video Production

With over 70 AI copyright lawsuits now pending — more than double the roughly 30 filed at the end of 2024 — creators can no longer treat AI video compliance as an afterthought. The good news: staying on the right side of the law is less about avoiding AI entirely and more about controlling how you use it.

Secure copyright through substantial human creative control. The Copyright Office has concluded that prompts alone don't provide sufficient human control to make users authors of AI output, so your protection comes from what you do after generation. Editing, selecting, and arranging AI-generated clips in a minimally creative way can establish protectable human authorship — the Office has registered hundreds of hybrid works, with coverage extending to the human contribution only.

Never generate recognizable copyrighted characters or competing content. Disney, Universal, and Warner Bros. have all sued AI video generators over protected characters like Star Wars, Shrek, and Looney Tunes. Under the Supreme Court's Warhol v. Goldsmith decision, a recreation serving the same entertainment purpose as the original weighs heavily against fair use, and the Copyright Office's May 2025 report confirms fair use doesn't shield outputs that compete with original works in their existing markets.

Disclaim AI-generated portions when registering. Failure to identify AI material in a copyright application led to a registration denial in the Théâtre D'opéra Spatial case. When filing, clearly separate your human contributions from machine-generated elements.

Vet each tool's licensing posture before adopting it. The industry is splitting into opt-out models like OpenAI's Sora — which requires copyright holders to block specific characters via individual requests — and licensing-based approaches like Bria's royalty system or Suno's planned 2026 licensed models. Your tool choice directly shapes your legal exposure, so check training-data practices and indemnification terms before committing to a platform.

Monitor the moving legal landscape. Key questions remain undecided, with the Andersen v. Stability AI trial scheduled for April 2027 and the New York Times v. OpenAI litigation consolidated into a multi-district proceeding. Treat today's guidance as a snapshot, not settled law.

For teams building AI video workflows, the same diligence applies to the agent skills and tools you chain together. SkillzBench is building a free directory that pins every skill to a commit SHA, runs license checks and security scans, and publishes a trust score — so you can see a skill's licensing posture before it touches your pipeline. You can join the waitlist now.

The bottom line: AI video creation is legal, but ownership and infringement risk depend on how much human creativity you add and what source material your tools draw on. Build both into your process from day one.

Automating a Legally Safer AI Video Workflow

Knowing the rules is one thing; applying them consistently across dozens of video projects is where teams slip. A single forgotten disclaimer or an unvetted AI tool can quietly undo an otherwise careful workflow.

The good news is that most legal hygiene is repetitive, which makes it a natural fit for automation. Agent skills — instruction packages that extend coding agents and chat-based AI assistants — can standardize your pipeline so the checklist runs itself. For example, skills can be set up to:

• Enforce disclosure steps, flagging any video that lacks the required AI-generated content disclaimers before it ships.
• Log human contributions — edits, arrangement, and selection — so you can document the creative control the U.S. Copyright Office looks for when evaluating authorship.
• Check each AI video tool's licensing posture and terms before it enters your stack, since tool choice directly affects your exposure.
• Review prompts and source material to catch references to protected characters before generation, not after.

The stakes are real. More than 70 AI copyright lawsuits are now pending, more than double the roughly 30 filed by the end of 2024, and film studios have entered the fray for the first time — Disney, Universal, and Warner Bros. all sued AI video generators in 2025. Anthropic's $1.5 billion settlement over pirated books, the largest payout in U.S. copyright history, shows how expensive sloppy data practices can become.

But automating a workflow introduces its own risk: which skills can you actually trust? Published research found that roughly 26% of agent skills carry at least one vulnerability, and there has been no reliable way to tell helpful skills from risky ones.

That's the gap SkillzBench is being built to close. It's a free, independent directory of agent skills that discovers skills from GitHub, pins every version to a commit SHA, and runs license checks and security scans before anything is listed. Each skill is benchmarked in a sandbox — tested with and without the skill on the same model — so you can see the measured Lift in percentage points rather than trusting marketing claims. A 0-100 trust score combines security, quality, license, maintenance, and community signals, and a free MCP connector lets you attach only the skills you've chosen to any agent.

Since the law here is explicitly a moving target — key fair-use questions remain undecided in cases like Thomson Reuters v. ROSS, and the Copyright Alliance notes these issues "will continue to develop" — a pipeline that re-scans and re-benchmarks as upstream skills change is exactly the kind of operational discipline that keeps your workflow compliant as rules shift.

SkillzBench is in pre-launch, and the waitlist is open now. Join it to be first in line when the directory, security scans, and benchmarks go live — and to turn your legal checklist into a workflow that runs itself.