How to Stop AI From Training on Your Content | Elacity
How to stop AI from training on your content: the new shared robots.txt rule, the named crawlers to block, edge enforcement, and why every opt-out is still a request.
How to Stop AI From Training on Your Content, and Where Opting Out Stops
To stop AI from training on your content, add a no-training rule to your site's robots.txt, keep blocking the named training crawlers, and enforce the block at your network edge. Do all three and you have stated your terms as clearly as the web allows today.
What you have not done is make those terms binding. Every step asks the crawler to comply, and your writing, photos and code stay usable by anyone who decides not to.
The new shared opt-out for AI training
On October 30, Google announced plans to support a shared AI training control in robots.txt, built on work from the IETF's AI Preferences working group. One rule would replace a separate instruction for every AI developer, and Google says it will read that opt-out as a block on Google-Extended unless you set Google-Extended yourself.
Microsoft said the same day that it will honor the shared training preference from February 1, 2027, and split training from grounding, the use of your page to answer a question in real time. Google's crawler documentation puts its own support in early 2027.
How to opt out of AI training today
A practical setup for a site you control:
- Add the shared rule as its own group in robots.txt: a line reading User-Agent: usage_ai-training, then Disallow: / on the next line. Providers that support it will read it on their own schedules.
- Keep the named rules in place. OpenAI says disallowing GPTBot indicates your content should not be used to train its foundation models, and Google-Extended still controls Google's training use until the shared rule takes over.
- Enforce at the edge if your host or CDN offers it. Cloudflare's own documentation states that robots.txt compliance is voluntary and points to its crawler controls to actually block selected bots.
- Test before you publish. A stray Disallow: / in the group that applies to every crawler removes you from ordinary search as well.
Why every opt-out is still a request
The standard behind robots.txt says it plainly: its rules are not a form of access authorization. The new working group is no different, since its charter lists technical enforcement of preferences as out of scope.
That gap has been tested. In 2025 Cloudflare reported that Perplexity was using stealth, undeclared crawlers to get around no-crawl directives. Cloudflare had already changed its default to block AI crawlers unless they pay, which puts enforcement with the network operator rather than with you.
Even a compliant provider promises less than it sounds like. Microsoft's own post says that neither allowing nor refusing guarantees inclusion, traffic, or payment. For work already collected, the fight moves to court: on November 2 a federal judge let a proposed class action proceed that accuses Runway AI of scraping YouTube users' videos to train its models, a case that will take years to settle what a robots.txt line could not.
We made the same point about license files in AI training data where the terms stayed in the README: terms that live beside an open copy depend on the reader's goodwill.
What ownership adds that a preference cannot
An opt-out tells a reader what you want. Sealing the work, so the key goes only to those who hold the rights, is meant to decide what everyone else can open. That is the gap Elacity aims to narrow, so the people who make the work own it as an asset instead of supplying it as free raw material.
1. The work carries its terms
Through Elacity dDRM, a song, dataset, image library or body of work is designed to become a Wealth Capsule: an encrypted good with its terms recorded alongside it, covering who may open it and at what price. A model builder that wants it would acquire rights under those terms, not a raw copy.
2. The key answers to rights, not to the crawler
Copying the file is not meant to be enough. The key that opens a Wealth Capsule is designed to be split across a 2-of-3 quorum of separate machines, each re-checking the holder's on-chain rights before releasing its share. The longer-term aim is to release keys only to a device that can prove what software it runs. That path is not switched on yet, so read it as the design, not a guarantee. And once a rights holder opens the work, what happens next rests on their licence, and in time on hardware checks where the device supports them.
3. Your data starts on your own machine
Elacity dDRM is designed to plug into ElastOS, the open runtime Elacity helps build, where your data starts on your own machine and reaches the cloud only when you choose to publish it. In ElastOS, an app can't make its own permission: it asks, and the Runtime mints a capability token, then checks it before the app can act. Capabilities, not access control lists: authority starts at zero and is granted narrowly.
An honest limit: a page you want everyone to read cannot also be sealed. Keep robots.txt for your public pages, and seal the work you sell or license, such as datasets, full-resolution images, archives and models. Our decentralized DRM explainer covers who holds the key, and more on the rules being written for AI lives under Ecosystem & Governance.
Quick answers
Does robots.txt stop AI training?
No. It states a preference that compliant crawlers honor, and the standard itself says it is not access control.
Does opting out remove my work from models already trained?
No. Microsoft describes its preferences as covering future training and grounding; content already collected is a question for licensing or the courts.
Should I block AI search answers too?
That is a separate choice. Blocking grounding keeps your pages out of generated answers, and Microsoft warns it also reduces your visibility there.
State your terms, then hold them
Add the shared rule this week; it costs nothing and puts your preference on record. Then treat it as a sign on the door, not a lock. Follow Elacity on X for the next guide on owning the work AI wants to learn from.