- Releasing the source code under the Apache license
- Algorithm components and digital repository scale
- Transparency tool and account restriction checks
- Researcher participation and code pull requests
- Safety exceptions, political context, and integration
- Frequently asked questions
Releasing the source code under the Apache license
Social media platform X announced on Thursday an unprecedented expansion of its open-source codebase by publishing the complete source code for its default For You feed algorithm and core post-ranking and recommendation engine on global platform GitHub under the Apache 2.0 open-source license. This step was accompanied by the launch of an advanced in-app transparency tool allowing users to check whether their accounts or tweets have undergone any restriction or classification actions by the platform’s systems.
This initiative gives developers and researchers worldwide direct access to detailed code instructions determining how tweets are fetched, ranked, and distributed in the user interface, alongside understanding weights and criteria adopted to process various engagement signals, representing a major shift toward digital transparency in social network management.
Algorithm components and digital repository scale
Keith Coleman, vice president of product at X, explained in press statements that the newly available code repository has become roughly 10 to 15 times larger compared to previous partial releases published by the platform. The repository includes model configuration details, filtering systems, and criteria for classifying problematic content or material violating general platform rules and policies.
Coleman emphasized that the published code enables specialists to run evaluation and scoring systems outside company servers and understand the mathematical mechanics governing content display for hundreds of millions of users, noting that the programming community will be surprised by the scale of transparency and boldness in making these vital systems available after years of being closely guarded trade secrets among major social platforms.
Transparency tool and account restriction checks
Alongside publishing the code, X began testing a dedicated transparency tool appearing inside application settings under the title Behind the Scenes. This feature allows users who have posted ten times or more over the past month to download an aggregated JSON data file detailing precisely whether any evaluation labels or restrictions were applied to their accounts or posts during the past calendar month.
Non-technical users can leverage this feature by inputting the extracted data file into any artificial intelligence large language model and linking it to the X repository URL on GitHub to obtain an immediate and simplified natural language explanation of how the platform’s algorithm classifies their accounts and the reasons behind the visibility or decline of their posts. The feature began in an initial pilot phase targeting active accounts that are at least one full year old.
Researcher participation and code pull requests
Before the official launch, X shared its source code with independent external researchers specializing in recommendation systems, who successfully trained and ran the platform’s Phoenix evaluation system using the open code, marking a prominent milestone in verifying the initiative’s reliability. The company announced it will open doors for developers worldwide to submit code modification and improvement proposals for the algorithm via GitHub for review and integration of useful ones into the platform’s live system.
Coleman expressed enthusiasm for this step, stating: “Our vision is to empower the public to review and evaluate the fairness of the software system, so that the algorithm is not only visible to the public, but developed and audited in collaboration with the digital community itself to build a platform that enjoys everyone’s trust.”
Safety exceptions, political context, and integration
Despite the wide openness, X excluded certain sensitive security systems from the published code, most notably artificial intelligence models like Grok used to predict rule violations, aiming to protect the platform from malicious actors who might exploit that information to bypass policies and flood the network with spam and fraudulent messages.
These developments come in the context of addressing ongoing historical concerns regarding the impact of recommendation algorithms on elections and information dissemination, and previous allegations of shadowbanning accounts. They also coincided with the announcement of the completion of X’s integration with SpaceX and periodic public updates on monthly active user counts.
Frequently asked questions
Question: What open-source license was used to publish X’s algorithm?
Answer: The algorithm was published on GitHub under the Apache 2.0 open-source license for developers and researchers.
Question: How does the Behind the Scenes feature help users check their accounts?
Answer: It allows downloading a data file showing whether their accounts or posts underwent restrictive classifications or were affected by recommendation systems over the past month.
Question: What parts were excluded from the open code for safety reasons?
Answer: Artificial intelligence models like Grok used for predicting violations were excluded to prevent exploitation by spammers and malicious actors.
