Report: Tumblr's Owner Strikes Deals with OpenAI and Midjourney for Training Data

 

According to a report by 404 Media, the owner of Tumblr and WordPress.com, Automattic, is reportedly in discussions with AI firms Midjourney and OpenAI to supply training data extracted from user posts. Citing an anonymous source within the company, the report suggests that agreements between Automattic and the AI companies are on the verge of being finalized. Speculation has circulated on Tumblr in recent days, hinting at a potential partnership with Midjourney that could offer a fresh revenue source for the platform.

404 Media's report suggests that Automattic intends to introduce a new feature on Wednesday enabling users to opt-out of data sharing with third parties, including AI firms. However, internal communications imply that the company inadvertently collected an "initial data dump" encompassing all public post content from Tumblr between 2014 and 2023, some of which may not have been publicly visible. The fate of this data and whether any has been shared with Midjourney and OpenAI remains uncertain.

Both OpenAI and Midjourney have not responded promptly to The Verge's request for comments. Automattic referred to a public statement released on Tuesday in response to 404's report. Titled "Protecting User Choice," the statement acknowledges partnerships with undisclosed AI companies. It asserts that Automattic presently blocks major AI platform crawlers by default and shares only public content hosted on WordPress.com and Tumblr from sites that haven't opted out. Additionally, it emphasizes collaboration with select AI companies aligned with user concerns regarding attribution, opt-outs, and control.