The AI Legal Liability Question Hollywood Is Not Talking About Loud Enough

The liability question that will define what AI creators can actually build, and who pays when something goes wrong

There’s a panel topic from the AI on the Lot conference that didn’t get nearly the attention it deserved.

Not because it was buried in a side room. It had serious lawyers on the stage and serious questions on the table. But compared to Amazon’s greenlight announcements and Paul Schrader’s predictions about silicon based movie stars, it didn’t have the same headline energy.

It should have.

The panel was about Section 230. And if you are building anything with AI tools right now, commercially, creatively, or both, this is a legal question that determines how exposed you may be when something goes wrong.

Here’s what you need to understand...

What Section 230 Actually Is

In 1996, Congress passed a law that essentially said internet platforms are not automatically responsible for what their users do on them. If someone posts something defamatory on a message board, the platform usually is not treated as the speaker. The user is. That protection is Section 230 of the Communications Decency Act.

It’s the reason the modern internet exists the way it does. YouTube, Reddit, Facebook, none of them could have scaled in the same way without it. The liability would have been impossible to absorb.

For thirty years, that shield shaped the internet.

Now the question is whether it extends to AI generated outputs. And the answer coming out of AI on the Lot was not that the issue is fully settled. It is not. But the legal argument against blanket Section 230 protection for AI is getting stronger.

Why AI Is Different

The argument for extending Section 230 to AI goes something like this: an AI model is just a tool, users prompt it, users are responsible for the outputs. The platform is passive.

The problem is that AI is not passive in the way a message board is passive. A message board hosts what a user types. An AI model actively shapes, generates, and contributes to the output. That is a fundamentally different relationship.

Aaron Rubin, a partner at Morrison & Foerster, made this point plainly at the conference. In the most basic case, a user types a prompt, a model returns a response, it is, in his words, “fairly hard to argue that Section 230 applies.” The AI model has contributed to that output. That is not pure user-generated content, which is what 230 was designed to protect.

Luke Arrigoni, CEO of Loti, a company that specializes in digital security and likeness protection, went further. As reported from the panel, his view was that platforms should not expect immunity for the things they help generate.

This is not a theoretical debate. Courts and lawmakers are already moving.

So Who Is Actually Liable?

This is where it gets complicated, and where it matters most to you as a creator.

If an AI platform generates something unlawful, a defamatory output, an unauthorized likeness, or other legally actionable material, Section 230 may not shield the platform in the same way it shields a traditional website hosting a user’s post.

That is actually good news on one level. It creates pressure on the tools themselves to be built more carefully.

But it does not necessarily protect you. The platform’s liability and your liability are not mutually exclusive. If you used the tool, directed the output, selected it, edited it, published the result, submitted it, or profited from it, you may be on the hook too.

Copyright adds another layer because copyright is not mainly a Section 230 issue. Section 230 does not generally protect platforms from intellectual property claims. So when people talk about AI training data, copyrighted works, fair use, licensing, and whether outputs are derivative, they are talking about a separate legal risk category sitting alongside the Section 230 question.

The skeptical view of AI in Hollywood is precisely this: that these models may have been trained on copyrighted material, and that every commercial output raises questions about what the creator, the platform, the studio, or the distributor can actually prove.

Whether that view holds up legally is still being tested. But the direction of travel is clear: creators are going to be asked harder questions about what tools they used, what those tools were trained on, what rights were cleared, and whether there is a clean chain of creation behind the work.

The backlash around Jorge Gutierrez’s Punky Duck project is a good example of how quickly those questions can become radioactive. The concern was not just one thing. It was authorship, training data, consent, creative labor, Amazon, artists, and whether anyone could really explain where the work came from and who had been protected.

That is the environment AI creators are walking into.

The Legislation Already Moving

Two bills are worth knowing by name.

The NO FAKES Act, championed by SAG AFTRA and others in Hollywood, specifically addresses AI-generated likenesses. If enacted, it would create a federal right around voice and visual likeness and allow action against those who knowingly create, distribute, or profit from unauthorized digital replicas.

The Take It Down Act is further along. It is already federal law. It targets nonconsensual intimate imagery, including AI-generated deepfakes, and requires covered platforms to remove validly reported material and known identical copies within a short window. This one has bipartisan support, which in the current environment is notable.

On Section 230 more broadly, Rubin noted at the conference that there is “unusual bipartisan alignment” against extending its protections too easily right now. People on the left don’t like it for one set of reasons. People on the right don’t like it for different reasons. The result is the same: the shield is under pressure.

One moment from the panel that landed with the room, a moderator pointed out, mid discussion, that the conference itself was sponsored by Amazon. The lawyer on stage laughed and said, “Exactly.”

That tension is not going away.

Put This Into Practice

Before you ship any AI-generated content commercially, ask three questions your legal counsel will eventually ask you anyway.

First: Do you know how the model you used was trained, and whether the training data was licensed? Most creators don’t. That gap is a liability exposure.

Second: Does your contract with a studio, platform, brand, or employer include an indemnification clause that places the legal risk on you if the AI output is challenged? Read it. These clauses are increasingly standard and increasingly aggressive.

Third: Does your work involve any real person’s likeness, voice, identity, performance style, or recognizable persona, even incidentally? The NO FAKES Act is aimed directly at unauthorized digital replicas, and the boundaries around incidental use have not been fully tested in court yet.

The conference had a panel on this. Most of the hallway conversation was about greenlights and box office, and what model produced which texture. That’s understandable. But the legal layer is the one that will determine, when a deal goes sideways, who pays for it.

Know it before you need it.