OpenAI and Midjourney × Automattic Inc.
In February 2024, it was reported that Automattic, the parent company of Tumblr and WordPress.com, was finalizing a deal to sell public user data to OpenAI and Midjourney for AI training. [25, 28] The announcement was met with user concern, particularly after internal documents suggested that data not intended for sharing, such as private posts and content from deleted blogs, had been compiled. [26, 27] In response, Automattic published a statement confirming it works with select AI companies and introduced an opt-out toggle for users. [26]
The asset
A deal to sell a corpus of user-generated public content from Automattic's properties, Tumblr and WordPress.com, to be used for training AI models.
The record
- Seller status
- Going concern
- Sector
- social network / blogging platform
- Modalities
- text, images
- Jurisdiction
- US
- Personal data
- Users are opted-in by default to having their data shared, but the company introduced a setting to allow users to opt-out. Reports indicated that the initial dataset compiled for the deal included private posts, content from deleted accounts, and other data not intended for sharing.
What it means if you hold data
This deal illustrates the complex PII and user consent issues involved in licensing UGC for AI training. The decision to make data sharing an opt-out rather than opt-in feature, and the reported inclusion of private data, highlights the potential for conflict between platform operators and their user communities over data rights.