Stay discoverable in search while disallowing AI training | Hacker News Reader

Stay discoverable in search while disallowing AI training

blog.cloudflare.com

87 pointsdjfergus50 comments

I feel like most of these schemes to categorise data as "public but not really" are ultimately doomed to failure. Even if you could trust every AI company in the world to respect these terms, is there anything stopping someone else indexing the data and selling them the information? I know there's copyright law but they're apparently ignoring that anyway.

Ultimately this reminds me of those really early social media profiles (before people understood privacy settings if they even existed) which would say "If you're not my friend you're not allowed to read this page".

If you don't want your content to end up in some database/archive don't publish it for the whole world to see.

It's also a way to literally advertise to AI companies that you have some data worth plundering in an increasingly dead and sloppy internet that has diminishing returns for training.

> If you don't want your content to end up in some database/archive don't publish it for the whole world to see.

This principle somewhat reminds me of the line that "If you're not paying for the product, you are the product", and it seems to me similarly misleading - my data gets harvested and sold by companies with which I have non-paying relationships and by companies I have to pay for things (I am made the product in both cases). As you note - the AI companies are ignoring copyright law and pirating everything that seems useful to them regardless of whether it was published for free access.

The potential externalities here are troubling.

https://vbuckenham.com/blog/how-to-find-things-online/

"Cloudflare classifies bots by behavior, and a single bot can exhibit more than one behavior."

Is that really true

CF classifies anyone not using a popular browser with Javascript enabled as a "bot"

CF fingerprints www users

As an example, look at CF's Permissions-Policy HTTP response header on a site with CF "bot protection", i.e., the "checking your browser" CAPTCHA nonsense (challenges.cloudflare.com). Then look at IA's Permissions-Policy response header. One CDN is advertiser-focused, the other is user-focused

IA = Internet Archive

As far as I can tell, after months of fighting being DDoSed by Anthropic and OpenAI across 50+ sites - Cloudflare also allows what it considers "good bots" through all of your bot blocking rules, with no option to turn this off unless you pay them money.

The "good bots" also just happen to be from companies that pay cloudflare a lot of money, I imagine.

Goodness Tokens

How did you dertmine it was those two companies? Also did you disallow them in robots.txt?

No, I definitely spent months fighting off bots across multiple hosts and never set up a robots.txt file anywhere nor did I look at the analytics dashboard on cloudflare.

I'm asking sincerely, how do you get to the conclusion that it's those two?

To that point, Cloudflare quite literally tells you the source of the bot traffic on your sites that use it as a WAF! And what percentage they are allowing through, despite you setting rules specifically to block them by name.

It’s baffling.

CF classifies anyone not using a popular browser with Javascript enabled as a "bot"

This is my biggest complaint about CF. They are implicitly supporting user-agent discrimination in favour of Big Browser, instead of discriminating on actual behaviour.

...and of course there are already companies running tons of VMs with "officially sanctioned" browser + OS stacks, that can get past all these "protections", for a fee.

"AI bots" is the newest boogeyman they came up with to take away freedom.

11,9,10

"Accountable" is just a fancy word for "pinky promise, but with a label." Nothing stops the data from ending up in a training run once it's already been fetched.

...and the only way to stop[1] that is by effectively DRM'ing everything, which is a level of dystopia that I don't think even Stallman ever anticipated, nor do I want to happen.

[1] Analog hole and other workarounds aside, naturally.

> "pinky promise, but with a label."

and that label is "pending litigation"

I didn't know websites could opt out of providing data to Google's AI training. Looks Google added support for this via 'Google-Extended' in robots.txt back in 2023:

https://blog.google/innovation-and-ai/products/an-update-on-...

doc: https://developers.google.com/crawling/docs/crawlers-fetcher...

The irony is that search engines are AI companies now. Telling them 'index me for search but don't train your models' is asking them to split a brain that’s already fully merged.

Exactly. I don’t get this. You may decide to not train but your content will still show up in search.

Even looking for companies which supply services is now far better on AI chats than Google. For me being visible in AI training is going to be more important than search in the next year or two.

If I was a big AI company I'd certainly be tempted to make sure that anyone who excluded themselves from "AI training" also got themselves excluded from AI results.

I maintain a cloud IP ranges database, and I'm going to test this out.

I have my doubts, though. A formal title like "Accountable" (capitalized) sounds deliberate, but I can't help imagining the renewal email:

"Hey, want to renew your Accountable™ license? Just pinky promise again that you use your IPs for what you say you do."

I wonder if protocols like Web Bot Auth [1] will see wider adoption. At least as a supported mechanism for those bots which identify themselves. The rest probably still have to be treated with Anubis. In my free time I've recently been experimenting with a Web Bot Auth implementation as an Envoy dynamic module [2] to have a way to define some additional policies for the traffic from bots.

[1] https://datatracker.ietf.org/doc/draft-ietf-webbotauth-https... [2] https://github.com/michalskalski/envoy-web-bot-auth

I'm playing around with it for my MCP hiring protocol ojcp[1] and it seems to work very well for signing attestations at the header level.

[1] https://github.com/ojcp-org/ojcp

Thanks for sharing, it is interesting use case

> Accountable mixed-use crawlers remain allowed for search. Every other training crawler is blocked, including the training-only crawlers run by Amazon, Anthropic, Meta, and OpenAI — blocking those does not affect search.

> We also categorize the relevant crawlers from Amazon, Anthropic, Meta, and OpenAI as Accountable. These organizations separate their Search and Training crawlers, so Cloudflare can block the Training crawler without affecting search.

I find it difficult to trust that either Meta or OpenAI would use their separate search and training crawlers only for the respective purposes. Their pinky promises have no value, IMO. Both companies are premised on deceptive behaviors.

What does this really mean though? You can use an LLM to search.

In a world where people's searches are already mostly answered by AI, is there any point to disallow the training?

I could totally see a future where Google just stops showing you links to actual web pages altogether and just gives you their chatbot.

I wouldn’t trust an AI company to honor this as far as I could throw them

"Apple, Google, and Microsoft honor or have committed (in a specified time frame) to honor this setting."

Specified Time Frame ;)

"infinity is a timeframe!"

there have been a few links here on hn about content protection based on Markov’s chain. It might be interesting for Cloudflare to add damage for the scraper that tries to query the page, and not just blocking them.

“Pre-label your most valuable data for us and we promise to not make it too obvious that we’re training on it”

Cloudflare, enabling the problem and the solution since, how long has it been?

(checks watch)

17 years

Fun lava lamp story, though

This seems useful for my site. I don't really want competitors' AI systems use our research data to train their models.

what weirds me out is the analytics

theres no way my index.html page with nothing is getting 10000 hits a day

wtf?

7 per minute.

I use my high school’s website to test Internet connectivity bc the domain is short and they don’t do a TLS redirect (making it easy to detect WiFi portals).

neverssl.com

I use this too, but my high school url is 8 characters. neverssl.com is 12. :)

Wish the admin never added the SSL redirect, its literally the namesake lol

I just want basically captive.apple.com with a shorter domain and the webserver not even listening on port 443 at all.

Surprised someone hasn't made this yet, it only requires one spare public IP.

They bullshit big time on that, I have caught them not once.

For a site with proven visitors 100,000 per month CF tell me they saved 90,000 over 1,000,000 visitors (rough numbers).

Sure they know large numbers make people feel good.

If this admin is serious about AI growth they’d make anti-scrapping illegal.

I don't think this admin even knows what scraping is. Unless it's "scraping the bottom of the barrel", something they're very familiar with.

Just make user-agent-discrimination illegal.

What does this new setting actually do? Does it block their IP ranges too, I hope? As if Meta, for instance, is actually going to respect Accountable, via themselves or their partners, quite frankly is eyebrow raising at best.

Reply on news.ycombinator.com