AI Content
Risk Briefing

Does Runway's commercial license actually protect your brand?

When a brand or agency pays for a commercial subscription to an AI video tool, the assumption is that the tool company stands behind the outputs. For one major tool, that assumption is correct. For the tools producing the most creative AI video in the market today, it is not.

Runway's Terms of Service require users to indemnify Runway against losses related to copyright infringement. The protection flows in the wrong direction. If a rights holder comes after your brand for content produced with Runway, Runway will not defend you, will not cover damages, and may itself become a co-defendant — which increases your exposure, not reduces it.

In July 2024, leaked internal documents revealed that Runway's Gen-3 model was trained on YouTube videos downloaded without permission from Disney, Netflix, Pixar, and Sony — using proxy software specifically to avoid detection. Runway provides no contractual protection to the brands using these outputs commercially.

Adobe Firefly's indemnification is real, and it is the only tool in the market offering it at scale. The trade-off: Firefly's outputs are, by professional creative standards, limited. The highest-quality AI video — the work that makes brands take notice — is produced with Runway, Sora, Kling, and similar tools. SI8's Safe Lane process exists precisely in this gap: documented vetting that gives brands the defensibility that the tool companies don't provide.

YouTube, TikTok, Meta, Netflix: what each platform actually requires for AI video

A brand running a single AI video campaign across major platforms must now navigate four structurally different compliance regimes — with different triggers, different enforcement mechanisms, different penalties, and different update cadences. There is no single workflow that covers all of them.

The Interactive Advertising Bureau launched a quarterly Video Compliance Brief series in 2025 specifically because the cross-platform complexity had become unmanageable. Their characterization of the current environment: a "risk-based, materiality-driven approach" is now required — meaning no single bright-line rule works across all platforms.

Layered on top of platform rules: in 2025 alone, US state legislators introduced over 1,100 AI-related bills. Approximately 100 were enacted or proposed into final rules, creating — in the words of legal analysts — "a patchwork of 50 different regulatory regimes."

Why "I paid for the tool" isn't a legal defense

The most common assumption brands make when using AI-generated content commercially: "we have a paid commercial subscription, so we're covered." The US Copyright Office addressed this assumption directly in May 2025. The answer is not what brands want to hear.

The Copyright Office's third and final AI and copyright report, published May 9, 2025, states explicitly that use of copyright-protected materials for AI model training is insufficient on its own to justify fair use. More critically: the knowing use of illegally obtained training data weighs against a fair-use defense — meaning if a tool was trained on stolen content, the downstream commercial users of that tool are relying on a legally compromised foundation.

In 2025, a US court rejected the transformativeness argument for commercial AI outputs — the idea that because AI-generated content doesn't contain the original work verbatim, it's not infringement. The court found that transformation does not occur simply because the copyrighted content does not appear verbatim in the output. This dismantles the most common secondary assumption brands make.

The pace of litigation makes this a now problem, not a future one. Active copyright infringement cases against AI companies more than doubled in 2025, from approximately 30 to over 70. Three of Hollywood's five major studios have now sued Midjourney. The RIAA sued Suno and Udio; Suno settled and committed to retiring its current model — meaning brands with Suno-generated audio in active campaigns are using content from a model that was legally indefensible enough that its creators agreed to shut it down.

Under vicarious liability principles, a brand that commissioned AI-generated content and profited from it can be held liable even if an agency produced the infringing work. Major IP law firms are now advising brand owners to amend agency contracts to add explicit AI-specific IP warranties and indemnification clauses — precisely because standard contracts do not contain them.

AI content disclosure deadlines: what's changing in 2026 and what brands need to do now

The legal risk of using AI-generated content commercially is not hypothetical or distant. Several jurisdictions have already passed binding requirements, and multiple hard deadlines are approaching in 2026. Brands building AI content into campaigns now are building against a compliance calendar that is actively tightening.

Beyond these hard deadlines, the pace of state-level legislation in the US is accelerating regardless of federal action. Over 1,100 AI-related bills were introduced in US state legislatures in 2025 alone. Tennessee's ELVIS Act (2024) extended right-of-publicity protections specifically to AI-generated voice clones. California's AB 2602 (2024) bans use of digital replicas of deceased performers without estate consent. Massachusetts is advancing mandatory AI disclosure requirements for any generative content within the state.

SI8's Safe Lane process is designed to produce the documentation that satisfies these requirements at the point of production — not as a retroactive compliance exercise. Every Rights Package includes the tool provenance log, authorship declaration, commercial license confirmation, and disclosure-ready metadata that platforms and regulators are converging toward as the standard.