Bluesky evaluates the development of a approval mechanism on how to use user data in artificial intelligence training. Jay Graber, CEO of the company, said in a statement at the SXSW 2025 conference in Austin, that this new system will allow users to check whether their content will be included in artificial intelligence models.
In recent years, the rapid development of artificial intelligence technologies has raised the use of public content on social media platforms by large language models and productive artificial intelligence systems. Bluesky aims to offer more control to its users as a decentralized social network. However, the platform’s open structure allows third parties to use user data in artificial intelligence training. In a study by 404 Media last year, it was found that 1 million posts shared in Bluesky were used as artificial intelligence training data on the Huging Face platform.
Bluesky’s rival, X (formerly called Twitter), is among the platforms that include users’ shares in the artificial intelligence development process. The X uses the content on the platform to train Grok’s chat boat directly integrated with the XAI, the artificial intelligence attempt of Elon Musk. In the autumn of 2024, with the amendment of the company’s privacy policy, third parties were allowed to use the shares in X to educate artificial intelligence models.
These developments have led to users who were not satisfied with the policies of X to turn to alternative platforms. Especially during the US elections, Elon Musk’s rapprochement with the Trump administration and the criticism of X for disinformation made a significant increase in Bluesky’s number of users. The company reached more than 32 million users in a short period of two years and caught a huge momentum.
Bluesky CEO Jay Graber said that the company aims to prioritize user preferences in artificial intelligence training processes. According to Graber, the system to be developed can work in a similar way to the Robots.txt file that determines whether the websites will be scanned by search engines.
“Although websites are generally open to the public, many search engines respect the rules specified in the robots.txt file. We think that a similar mechanism can be applied for social media platforms. Grart Graber said that a system in which users can determine whether their contents will be used by artificial intelligence systems should be developed.
According to the proposal to the community view on Github, users will be able to choose whether their shares will be used in artificial intelligence training at the account level or on certain posts. Then, other companies will be requested to respect these preferences.
Graber said, “We need the support of users, companies and regulatory institutions in order to implement this system. However, with the right framework, we can ensure that users have more control over their data. ”
Bluesky can be an example for other social media platforms
This step by Bluesky strengthens the calls that social media platforms should adopt more transparent policies on how to use user data by artificial intelligence. Today, more and more companies benefit from social media content to educate artificial intelligence models. Therefore, the spread of mechanisms based on user approval is considered as an important development in terms of digital rights.