Tumblr owner is selling user posts to OpenAI and Midjourney

If you noticed Tumblr users panicking this week over AI training rumors, it turns out they had good reason.

According to a report from 404 Media citing an internal source, Automattic, the parent company of Tumblr and WordPress.com, is preparing to close deals with OpenAI and Midjourney. The agreements would hand over user-generated posts to train generative AI models, and internal chatter suggests the contracts are imminent.

Rumors had already been swirling across Tumblr over the past week, with users speculating that a Midjourney partnership was brewing as a desperate bid for new revenue.

Scraped archives and an accidental leak

According to the report, Automattic plans to launch a new setting on Wednesday that lets users opt out of data sharing with third parties, including AI firms.

However, internal posts revealed that the company already compiled an initial data dump covering all public Tumblr posts published between 2014 and 2023. Worse, that initial scrape apparently included private content that was never meant to be publicly visible on blogs.

It remains unclear what Automattic did with that specific dump, or whether any part of it has already landed on Midjourney or OpenAI servers. Neither AI firm immediately responded to requests for comment.

Automattic later pointed to a blog post titled "Protecting User Choice," confirming it is working directly with select AI companies. The company maintained that it blocks major crawlers by default and will only share public content from accounts that have not opted out, provided partners respect attribution and control.

Monetization woes meet creative pushback

Selling community data to AI labs has quickly become standard playbook. Reddit recently landed a reported $60 million annual agreement with Google, while Shutterstock licensed its vast photo catalog to OpenAI.

Yet Tumblr is a very different beast. Its base consists heavily of digital artists, fan writers, and indie creators who are fiercely protective of their work. DeviantArt saw intense user revolt when it pushed into AI, and Tumblr is walking right into that same buzzsaw.

Automattic acquired Tumblr from Verizon in 2019 but has consistently struggled to turn a profit from it, even announcing last year that it would scale back its ambitions for the microblogging site. Monetizing ten years of user posts feels like a shortcut to cash for an asset that has otherwise failed to deliver.

Why it matters

For Tumblr and WordPress users, the immediate takeaway is clear: check your account settings this Wednesday and flip the opt-out switch if you do not want your posts feeding AI models. Still, the revelation that private posts were scooped up in an internal scrape means opting out now may feel like locking the barn door after the horse has bolted.

For the AI landscape, the move underlines how scarce human-generated training data has become. As legal risks mount around unapproved web scraping, AI vendors are racing to secure legitimate pipelines to creative archives. Tumblr represents a goldmine of niche art and conversational text that Midjourney and OpenAI crave.

We think Automattic is playing with fire. Courting AI partnerships to monetize a struggling platform makes business sense on paper, but alienating the very creative community that keeps Tumblr alive is a massive gamble. Mishandling private data during the scrape only deepens that trust deficit, and no licensing payout will fix a mass creator exodus.