• they/it

I do art, sometimes


mcc
@mcc

I have a complaint about Cohost. I understand my complaint is not trivial to address, and I don't expect change to happen quickly, but I want to see it addressed at some point. It's section 6 of the TOS. I've been complaining about this for two years and I am planning to keep bringing this up gently about once a year.

TOS provision 6 grants Cohost the company an unlimited and effectively unbounded right to do anything, including potentially commercial exploitation without compensation, with the content you post to this site. This is extremely normal for a social media site, and because websites like Cohost are technically complex it is widely considered impossible to run a site like this without a broad clause like this. The clause by itself is standard and unsurprising. What is not usual, and is the source of my worry, is that Cohost's IP grant does not have a termination clause of any kind. The last time I posted about this I wrote a long-ass philosophical post comparing every social site I could think of. This time I'm just gonna be blunt¹. I want to call attention to three things that happened since the last time I posted about this:

  1. It has begun to be standard for a declining social site to sell their post histories to "AI" companies as training data. Tumblr², Twitter³ and Reddit have all three done this— and all three did it notably after mass exoduses from their respective sites⁴.

  2. Stack Overflow, shockingly, after their own deal to sell post content for "AI", took the totally unprecedented step of not only disallowing people from removing their own content, but banning people from the site for attempting to use the existing delete/edit features to remove content⁵.

  3. The Cohost team this March posted a financial update stating that, at the time they posted it, they would be running out of funds completely in April, and detailing⁶ what exactly their plans would be if the site shut down. It appears that some combination of efforts has moved the danger out of visible range⁷. However, the fact remains the Cohost team has now acknowledged that the future shutting down of the site for financial reasons is a possibility.

The first two of these items matter because they show there is now an existing market, for a site which no longer sees their route to success being to gain new customers, to rugpull by selling all their users' content for mass commercial exploitation. Before when I posted about this I was imagining frankly mildly implausible scenarios like "Cohost or someone who got hold of their IP puts your posts and art in a book and sells it". There's not really a market for that. But there is now a market for buying posts for "training". The worst-case scenario is no longer hypothetical. It's a multibillion-dollar business.

The third item matters because (speaking for myself here) the reason I trust the Cohost team would not sell my shit to OpenAI (or whatever other worst-case scenario bizarro IP abuse) is because I know some of them personally and I trust them. If you read the March update above, you'll see the team repeatedly stressing how committed they are to ensuring user data and IP are protected in a shutdown scenario. However, if there actually is a real possibility of Anti-Software Software Club going out of business, this opens up possible futures where it goes out of business in a fashion where those people I trust at ASSCLUB, and who made the promises in the March update, are not in full control of what happens to assets like user IP.

The solution to this, I believe, is for Cohost to join sites like YouTube, Facebook, and Twitter and add a termination clause for the IP clause, probably activating on some natural boundary like a post being deleted or an account being deleted. The TOSes of those other sites show this can be done in a commercially non-disruptive way. If site architecture means that deletion takes time, you can mimic text like YouTube has⁸ that the content is removed in a "commercially reasonable" timeframe. If your backup process means you've got site backups with old user IP interleaved, you can mimic— again— YouTube, and have the IP grant acknowledge there may be content which is retained for technical reasons but no longer commercially utilized. It should be possible to fit the termination clause to whatever Cohost's actual site practices already are, and so I expect (with the admission I have not myself attempted to run a site like this one) that the only expense involved in fixing this problem would be the lawyer time required to write the clause itself. That is not nothing! But it is still doable.


¹ She said, at the end of the second long paragraph of a eight-paragraph post. I'm really bad at being terse. Sorry :(

² Tumblr participating in this is actually really alarming because, as I describe in my last attempt at this post, Tumblr is the one social media site that instead of a termination clause actually attempts to narrowly define what kinds of usage the site has rights to for user IP. When I wrote that post, I believed this meant Tumblr users didn't have to worry about unpaid commercial exploitation that is not part of the operation of Tumblr as users understood it 2007-2023. Apparently I was wrong! I guess the workaround for Tumblr seemingly flagrantly violating their own IP terms is that they don't sell the rights to the user posts, they sell access to the user posts, since OpenAI is just like looking at the bytes or something and after that some magic process that isn't covered by copyright happens.

What does the Tumblr deal mean for our discussion here? A doomerist way of looking at it would be that because "AI" now means corporations can do things like copy and make derivative works of copyrighted content without any laws applying at all, the IP terms of a site like Tumblr, Facebook or Cohost don't actually matter, the corporations will just do what they want to do. I'd take a different tack: The Tumblr situation is a weak signal that termination clauses are actually superior to complex Tumblr-style IP terms that attempt to nail down exactly how the IP will be used. Again I refer to YouTube's termination clause which unambiguously states that YouTube will not "distribute" videos that have been deleted; even if allowing OpenAI to "look at" user content in exchange for money is somehow fair under the Tumblr license, it would seem to unambiguously be outside YouTube's license to user-deleted content. (I say this is a "weak" signal because it actually is conceivable the Tumblr-OpenAI deal will at some point in some court turn out to have not been legal.)

³ This is a very recent development, like, last 48 hours recent. If you have not deleted your Twitter account yet I suggest clicking here and unchecking the "allow your posts… to be used with Grok" checkbox. I suggest doing this even for protected Twitter accounts.

By the way, if you want to do the same for Tumblr, you can find the instructions for opting out of Tumblr's AI scrape here; your content has probably already been sent to OpenAI, but Tumblr claims that if content is deleted after being scraped they will "advocate" that the people they sold it to delete it from their datasets, so…maybe… that's… real? If you want to opt out of the internal AI scrape for Facebook, meanwhile, then my recommendation is to delete your account as quickly as possible.

⁴ I stress this because if a site does something anti-user like sell user content rights to a third party, then users who are engaged might do things that inhibit the anti-user action, like deleting posts or unchecking the checkbox to opt out of the content scrape. On the other hand, if you've got a lot of users who have recently simply quit, those users will very possibly not ever find out their shit is getting sold or that they needed to log back in to uncheck a checkbox to avoid this.

⁵ I personally got banned from Stack Overflow for a bit for adding a single sentence to each of my existing posts stating I do not consent to my text being used for AI training. The email informing me of this was deeply asinine.

⁶ (While stressing that they did not, at the time, believe these plans would have to be enacted. I have no desire to be unfair to ASSCLUB here.)

⁷ See the most recent financial update posted yesterday.

⁸ CTRL-F "Duration of License". The YouTube termination clause is a mere two sentences.


You must log in to comment.

in reply to @mcc's post:

However, if there actually is a real possibility of Anti-Software Software Club going out of business, this opens up possible futures where it goes out of business in a fashion where those people I trust at ASSCLUB, and who made the promises in the March update, are not in full control of what happens to assets like user IP.

This is something I tried to explain to people so much regarding selling data, not just with Cohost but with the likes of Kagi or other Internet startups. It's so easy to say "I trust them to not sell my data to someone bad!" Do you trust them to know for sure whether someone buying the site is bad? Do you trust the BUYER to never do anything bad? Do you trust the buyer to never sell it to anyone bad, and do you trust their judgement in knowing whether a second buyer would ever betray that trust?

The only way to protect yourself with any of this is for a site to never gather your data in the first place (which works more for Kagi than Cohost, where user posts must be stored) or to have some kind of legal agreement in place, because it can't just be "I trust them to do the right thing", you have to trust a potential endless chain of others if selling the site is ever a possibility.

(Speaking of Kagi, I already made a post about how Kagi's AI uses Cohost posts as sources, and often uses them to give extremely incorrect answers, even using your deleted posts/pages: https://cohost.org/lori/post/5466692-i-m-typing-a-bunch-o )

One big, MASSIVE problem on top of all this is the arbitration clause (section 28). We (in the US at least) completely give up any right to sue Cohost in court. These have decades of precedent in the US of being upheld, even if you agreed to it as a preteen.

The short of it is that users have practically no legal recourse for anything Cohost might do to them.

probably activating on some natural boundary like an account being deleted.

currently, to delete your account you have to email staff and then they'll get around to deleting it eventually. idk how long these requests take but i imagine the process would not survive a mass exodus event. users gotta be able to delete their accounts without staff involvement

IP termination clauses that trigger on account termination are relatively rare (Twitter is the only site I'm aware of) and it's much better for the users if they trigger on content deletion (because it lets users withdraw individual works without having to give up the site entirely). So if it's also logistically easier for Cohost to align it to post deletion, because that means they can do it without building out the account deletion feature… well, win-win?

I guess I should make this update:

The Cohost shutdown announcement post appears to address my concern above:

When the read-only period concludes, we will delete all of your data from our servers without a backup. Even now we want to reiterate that we think “data brokerage” and other common practices of the software industry are inimical to who we are as people, and we would never consider selling your data to others or asserting any rights to stuff you posted under any circumstance.

This is of course not the way I would have preferred the problem to be resolved, but I think making a public statement like this puts some legal restriction on the holders of the Cohost assets even if something "weird" happens to the ownership of ASSC or its assets over the next year. I am, to reiterate, not a lawyer.