Speaking at the SXSW conference in Austin on Monday, Bluesky CEO Jay Graber said social networks are working on a user consent framework on how data can be used for generating AI.
The demand for AI training data means that new social networks need to think about AI policies, even though they don’t plan to train their own AI systems on user posts.
However, the public nature of Bluesky’s social networks has already allowed others to train AI systems with user content, as discovered last year when 404 media came across a dataset built from one million Blueky posts embracing.
Meanwhile, Bluesky’s competitor X is providing user posts to its sister company Xai to help train the AI chatbot Grok. Last fall, we changed our privacy policy to allow third parties to train AI with users’ X posts. The move helped encourage the departure of other users from X to Blueski following the US election that raised the status of Elon Musk within the Trump administration, the owner of X.
As a result, Bluesky’s open source, distributed X alternative has grown to over 32 million users in just two years.
Speaking on SXSW, Graber explained that the company has engaged with partners and developed a framework for user consent on whether or not to use data for generation AI.
“We really believe in the choices of users,” Graber said that users can specify how they use Blueky content.
“It could be similar to how you specify whether your website wants to be scraped by a search engine,” she continued.
“Search engines can cut websites whether they have this or not, because the website is open on the public internet. However, in general, this robots.txt file is respected by many search engines,” she said. “So there’s something to be widely adopted and used by users, businesses and regulators, to use this framework. But I think that’s something that can work here.”
Proposals currently on GitHub include obtaining user consent at the account or post level, and involve asking other companies to respect that setting.
“We’ve been working with others in the space who are concerned about how AI will affect the way data is displayed,” Graber added. “I think it’s a positive direction that should be taken.”