Bluesky proposes letting users indicate if their data can be used for AI training, web archiving, and more; critics see it as a reversal of its prior statements
Social network Bluesky recently published a proposal on GitHub outlining new options it could give users to indicate whether …
TechCrunchAnthony Ha
Context & Ripple Effects
Bluesky had previously said it had no intention of using user content to train AI tools, then emphasized that outside organizations—not the network—were responsible for honoring consent when a Bluesky-post dataset surfaced for ML research. The new GitHub proposal moves the issue from a stated platform position toward user-declared permissions.
That shift matters because it makes the boundary between public availability and permitted reuse an explicit product and protocol question, while critics read it against Bluesky’s earlier assurances.
First-order effects
Users could be offered a way to signal whether their data may be used for AI training, web archiving, and other specified purposes; the report describes a proposal, not a completed rollout.
Bluesky must reconcile the proposed controls with its earlier stance and explain what, if anything, changes for data access and reuse.
Second-order effects
Researchers, archivers, and AI developers using public Bluesky data would need to determine whether and how to recognize those user-set permissions if the proposal is adopted.
The proposal puts renewed pressure on Bluesky’s credibility after its prior position that outside organizations should respect user consent—a standard that is harder to rely on without shared enforcement mechanisms.
Third-order effects
If social platforms increasingly attach reuse preferences to public posts, consent metadata could become a practical layer in the contest over whether publicly accessible data is automatically available for model training.
The unresolved issue is enforceability: user-facing controls can clarify expectations, but their industry value depends on whether downstream collectors and model builders adopt compatible practices.
The trend: Platforms are moving from broad statements about public data toward more granular consent architecture for AI and other downstream reuse.
We put up a proposal that lays out a way for users to declare whether/how they want their data to be used by things like generative AI or public archives, check it out on github: — github.com/bluesky-soci...
The fact that most of the content we all created before the genAI wave is being used with no restrictions, simply because there was no way for us to anticipate this kind of usage, makes me mad. — This is really necessary and I'd like to see more of this in other platforms! [emb…
This is a good proposal, and it's weird to see people flaming BlueSky for it as though this is equivalent to them welcoming in AI scraping (rather than trying to add a consent signal to allow users to communicate preferences for the scraping that is already happening) [embedded p…
Gen AI companies are already scraping public data from across the web, and everything on Bluesky is public like a website is public. But in the history of the open web, standards like robots.txt emerged that most search engines came to respect. This is a proposal to create a ne…
Great proposal, I love putting some safeguards and clarified intent on part of the author! Nice as a good actor dev to have these quickly available on query. — However everyone should note that none of this legally binding and that any public data will be scraped and used by a…
Right now there is a discussion on GitHub on giving Bsky users more control over how their content is used for AI training. Right now there is no limitation to it, but there are now discussions of an opt out system that many scrapers respect (via robot.txt). — techcrunch.com/2…
It's great to see @bsky.app team considering real user control over how their data is used for #AI. In a time when nearly all #SocialMedia platforms ignore privacy concerns, this proactive approach is a breath of fresh air. More companies should follow suit! #PrivacyMatters #A…
Seems like an ok thing, I'd prefer the AI gen permission to be disallowed by default though. Lot of people probably misunderstanding this as “bsky is adding in AI” - not the case here. Issue is gen AI (third parties) is already scraping bsky, this would (or should) allow you to…
This is a solid proposal. Declarative preferences are easy though, and only part of the solution. Governance and enforcement will be the hard part. [embedded post]
I noticed a lot of people misinterpreting this, despite the clear language. These proposals would be set to “undeclared” by default, not “yes”. That's a big difference compared to other platforms that set it to “yes” by default. If you'd never change these values, everything s…
Overall I think this is a good proposal. Feels like robots.txt is a good analogy, insofar as ATProto is a public protocol, and this would provide a way to give potential scrapers instruction. — I'm curious why they categorized by use case rather than agent, but looking forward…
One of the things Jay laid out at SXSW earlier this week. Bluesky has a robots.txt-like proposal to allow users themselves to declare if they're okay if they're data is used for AI training. [embedded post]
Technically this doesn't seem like a bad idea but I feel as though it falls short of user expectations for protecting their data. A wide majority of users will not have any idea what most of these preferences are about, let alone that they exist, and as such it feels functionall…
Absent industry buy-in, it would be great to see some kind of standardized conversion from intents into a robots.txt file that PDS and Relay operators would need to perform so these signals are harder for companies to ignore. — I don't know if robots.txt has that kind of granul…
My worry is that it's already kind of difficult to get AI companies to pay attention to robots.txt. — Since this is a completely separate standard, I don't know that there's any reason for a company like OpenAI to prioritize looking at it. They can claim ignorance and keep doi…
A standard is a good step, but like the proposal says, this would only be a signal - not something that Bluesky would be able to enforce. — Have any AI companies looked at user intents yet? Have they indicated that they would respect it?
I think the main difference is that if a search engine crawls and publishes the results it is easy to check. On the other hand, it is practically impossible to prove that a gen ai model has been trained with specific information. Beyond that, the intention is good, hopefully it…
If you want things like third-party clients and read APIs, all data on the network can and will be scraped & processed by anyone with an internet connection, with zero limitations. This isn't a ‘good’ or ‘bad’ thing (see the web) - just an inherent consequence of the hyper-publi…