The social media platform X has revised its developer agreement to prohibit the use of its data for AI Training.
Highlights
- AI training now restricted: X has banned third-party use of its API or content to train or fine-tune foundation or frontier models—reversing its previous policy from 2023.
- Policy shift tied to xAI: Following X’s acquisition by Elon Musk’s xAI, the platform draws a hard line—retaining internal use of user data for training while cutting off external access.
- New liquidated damages clause: Developers scraping over 1 million posts per day face a fine of $15,000 per million posts, signaling a strict stance against automated data collection.
- User opt-out introduced: Users can now opt out of having their data used for AI training via web settings: Settings > Privacy > Grok. Mobile support is expected soon.
- GDPR scrutiny rising: European regulators are probing whether X’s AI-related data practices align with consent requirements under GDPR, potentially triggering future enforcement.
- Creative communities push back: Artists and writers are voicing concern over content licensing terms that may allow use of their posts in AI systems without permission or attribution.
- Jurisdictional questions raised: Legal disputes must be resolved in Tarrant County, Texas, despite X being headquartered in Austin—raising accessibility and legal concerns.
- Part of a bigger trend: Like Reddit and other platforms, X is locking down its data as training content becomes a prized AI asset in a highly competitive landscape.
- Implications for developers: The changes limit access to public data streams, challenge open AI development, and raise new legal and ethical questions about consent and data ownership.
A new clause added under the “Reverse Engineering and Other Restrictions” section now explicitly bans third-party developers from using X’s API or any platform content to train or fine-tune “foundation or frontier models”—terms commonly used to describe large-scale generative AI systems.
This marks a significant policy reversal. In late 2023, X permitted limited AI training with its data, but the new restrictions now reflect a shift toward a more closed ecosystem.
A Strategic Shift Following xAI Acquisition
The policy update comes in the wake of X’s acquisition by xAI, Elon Musk’s artificial intelligence venture.
Since X amended its privacy policy in 2023 to allow internal use of public user data for model training, the platform has reportedly used this content to develop tools like its chatbot Grok.
The latest changes formalize a clear boundary: while X retains the ability to train its own models using user data, it now restricts external developers from doing the same—an approach aligned with a broader industry trend toward data exclusivity.
Changes and Implications
Introduction of Liquidated Damages
X has implemented a liquidated damages clause aimed at deterring large-scale data scraping. Under this clause:
- Any entity accessing more than one million posts in a 24-hour period could face financial penalties.
- The specified fine is $15,000 per million posts scraped without authorization.
This policy signals a stronger stance against automated data collection, particularly from AI developers using such data to train large models.
User Opt-Out Options for AI Training
While X restricts external AI training, it continues using user data internally. However, users have the option to opt out of their data being used to train X’s AI tools, including Grok. The opt-out path is:
- Settings and Privacy > Privacy and Safety > Grok > Uncheck the Data Sharing box
Currently, this feature is only available via the web interface, with mobile support expected in future updates.
Regulatory Attention in the European Union
The platform’s data usage practices have attracted scrutiny in the EU, particularly under the General Data Protection Regulation (GDPR). Concerns have been raised about whether users gave adequate consent for their posts to be used in AI model training.
Regulatory bodies across multiple EU countries are reportedly investigating complaints, which could lead to future enforcement actions or policy changes.
Concerns Among Creative Professionals
Artists, writers, and other content creators have expressed concerns about the broad licensing terms in X’s user agreement. These terms grant X significant rights over user-generated content, which some fear could be repurposed for AI training without credit or compensation.
This has reignited debate over intellectual property rights in the age of generative AI, especially for those who rely on digital platforms for exposure and income.
Legal Jurisdiction Clauses Raise Questions
X’s terms specify that any legal disputes related to its platform must be handled in:
- The U.S. District Court for the Northern District of Texas, or
- State courts in Tarrant County, Texas
This has raised questions among legal experts and users alike, since X’s headquarters is based in Austin, more than 100 miles from the listed jurisdiction. Critics argue that this could pose accessibility challenges for users seeking legal resolution.
Industry Trend Toward Data Lockdown
X’s updated policies reflect a growing industry-wide trend where companies are moving to protect their data assets from being used to train competing AI systems. Similar actions have been taken by platforms.
- Reddit, which has restricted AI crawlers
- The Browser Company, which added terms preventing model training via its browser data
As large-scale generative AI models continue to evolve, access to high-quality training data has become a competitive advantage—and a growing point of tension between open development and proprietary control.
