Can Cloaking Protect Your Work from AI?

Last week, the Writers Guild of America East (WGAE) announced that it was partnering with the AI protection company Cloak to provide its members with access to the company’s services.

The idea is relatively simple. If you are a member of the WGAE, you have free access to the service as part of their Secure Script Registration program. Non-members who register their work with the WGAE can also access it for an additional $10 per script.

According to Cloak, their service takes a provided script and “cloaks” it in a way where it is still readable by humans, but not by AI bots. This means that if the file is fed through an AI system, either automatically or deliberately, the AI will not be able to parse the document.

Cloak says that their system defends both against direct text extraction and optical character recognition (OCR) techniques. They claim that it has proven effective against all modern AI models as well as automated summaries from Google Drive, Gmail, and Adobe Acrobat products.

So is this the solution that writers have been waiting for? The answer is probably no. While it does have great potential for certain uses, this isn’t a broad solution for writers, especially for those looking to share their work more broadly.

How Cloaking Works

In the big picture, the idea of cloaking is straightforward. The idea is to create a document that is readable by humans, but not by AI bots.

We’ve seen something similar to this with images. Visual artists have used AI “poisoning” techniques to cause AI bots to misrepresent their work. The idea is that humans see one thing, but AI bots see something else, causing AI to misinterpret the image and, potentially, generate incorrect work based on it.

With Cloak, the company doesn’t go into detail about how their technology works, other than to say it is patent-pending. However, the cloaking takes place inside a PDF and they offer multiple layers of protection.

One potential issue is compatibility. Their “deepest” protection is only compatible with Preview on macOS. However, their more basic protection is compatible with Chrome, Adobe, Edge, Outlook and Google Drive. That said, specialty software, such as script-ending software is not compatible with their service. It’s also unclear if it is compatible with screen readers and other assistive technologies, though Cloak says to “assume it is incompatible.”

And that is likely the case. If screen readers can read the text, then most likely AI bots would be able to as well.

The goal of cloaking is to ensure that AI bots can’t read the material. The primary benefit is to protect your work from being used to train AI systems or otherwise be ingested by AI models. However, many scriptwriters will also likely appreciate that agents, producers, and others can’t simply generate a summary of the script and are forced to actually read it.

But does that mean it’s a good tool for most writers? The answer depends on what your specific goals are.

The Major Limitations of Cloaking

The biggest and most obvious limitation of cloaking is that, in order to prevent AI bots from accessing the content, you have to prevent other bots from doing so as well.

This includes search engine bots, screen readers and others that make documents indexable, accessible and searchable. Cloaked content, at least when done like this, will be harder to find, navigate and read.

The technology also has a limitation with file format. Cloak can only work with PDFs and, even then, only with certain readers. While you could theoretically publish a cloaked document on the web, it would be limited in who could find it, access it and read it.

To be clear, there are situations where this tradeoff is worth it. The WGAE has found perhaps the perfect use case – sending a specific document (a script) to a specific person (an agent or producer) who is expecting it already. This one-to-one or one-to-few scenario is probably the best use of this approach.

It is also best for potentially sensitive or high-value content. Where it might be reasonable to ask an agent or a producer to use a different tool to read a script, it’s going to be a much more difficult ask of the public at large.

For better or worse, the internet was built on the idea of broad access to text. Trying to retreat from that now and hide text behind what is essentially an anti-bot digital rights management system is fighting not just against over 30 years of history, but human nature.

That’s not to say that it’s impossible. But it would require an Internet that is very different from the one we have now.

Bottom Line

In the end, this is a very interesting story. The technology Cloak is using is fascinating, and its partnership with the WGAE is probably the best use of it. If it’s going to work anywhere, it’s in protecting high‑value content when being shared with specific people.

But, even in that context, I worry that it may not work. Many actors, producers and agents receive a large number of scripts, and I can imagine that many have adopted AI tools to help parse them. I’ll be curious to see what impact cloaking a PDF has on the chances of the script being picked up.

But that is something of a devil’s choice. If you want your work to be the most accessible it can be to your target, whether that is one person or the world at large, you almost have to open it up to AI bots. If you protect it, you may limit your audience, but you may be able to prevent AI bots from accessing the content.

There’s no easy answer here and a lot of it will depend on if the use of this technology becomes standard practice in the industry. If it does, it will become like the Academy Screening Room, a necessary hassle that’s tolerated for the sake of protecting the films that it streams. It could also expand to other industries, including book and music publishing.

But if it doesn’t become a standard and causes any harm to the scripts it protects, it will likely fade away. As with most things, money and success will be the driving force.

So the real question won’t be answered for a long while. Perhaps after a few years we’ll know if the practice has been truly normalized or not. But, by then, AI will likely have evolved and changed; it’s equally possible that this approach will already be obsolete.