Docs: Tumblr staff compiled users' data in preparation for deals where parent company Automattic would sell the data to OpenAI and Midjourney to train AI tools
Tumblr and WordPress.com are preparing to sell user data to Midjourney and OpenAI, according to a source with internal knowledge …
404 MediaSamantha Cole
Context & Ripple Effects
Automattic acquired Tumblr from Verizon for less than $10 million and later folded the service more closely into its organization; by late 2023, most Tumblr staff had been shifted elsewhere at Automattic. The reported data-sale preparations put a new potential revenue use on assets from that integration.
The later plan to move Tumblr's backend onto WordPress underscores Automattic's effort to join the platforms operationally. In that context, the reported preparation of Tumblr and WordPress.com data for AI buyers matters as a possible extension of shared-platform economics.
First-order effects
If the contemplated arrangements proceed, Automattic would turn Tumblr and WordPress.com user data into a training-data product for OpenAI and Midjourney rather than keeping it solely within its publishing services.
Users of both services face a changed data-use posture, while OpenAI and Midjourney would gain access to a curated platform-data source under a commercial arrangement.
Second-order effects
The prospect of platform-to-model-provider data deals raises pressure on other hosting and social platforms to clarify whether, and on what terms, user content and associated data can be licensed for model training.
Closer operational integration between Tumblr and WordPress could make common data-governance and consent choices more consequential across both services, rather than leaving each platform's data practices isolated.
Third-order effects
This is a data point in the shift from user-generated content as a platform engagement asset to an asset Automattic acquired with Tumblr that may also be commercialized as AI training input.
If such arrangements become routine, the durable fault line will be the public-data permission boundary: whether platform terms, user expectations, and disclosure practices can support data monetization for generative AI.
The trend: Generative-AI developers are pushing platforms to commercialize accumulated user data, making data rights and consent a central platform-governance issue.
It's easy to stop displaying 1st party info if a user toggles an opt-out; much harder will be figuring out how to remove that data later after it's been sold off to train some other company's AI system. “Automattic did not respond to a question from 404 Media about whether it co…
Following the announcement of Reddit's deal to sell access to user data to Google for $60M for training AI, it looks like Tumblr & Wordpress are in the process of signing similar deals with OpenAI and Midjourney. Access to training data & GPUs is going to be key in AI wars. [imag…
a Tumblr staffer wrote that they “unfortunately” compiled data for the initial dump to MJ/OpenAI that included: - private posts on public blogs - posts on deleted or suspended blogs - unanswered asks - private answers - NSFW content and more https://www.404media.co/...
Also @automattic's privacy notice says they don't sell your personal data for money. Except apparently they sell your personal data in tumblr and Wordpress to LLM companies (OpenAI) for money. https://automattic.com/... [image]
Selling Wordpress data seems especially heinous. It's a platform for website development. They should not have any rights over the user content on sites built on Wordpress. https://www.404media.co/...
Update to this: instead of answering any specific questions, Automattic published a public statement where it says, “We are also working directly with select AI companies as long as their plans align with what our community cares about” https://automattic.com/...
Controversial opinion: if we end up in a system where you can only train good AI models on $$ licensed data, there's a massive risk of concentration of power. It might not be the users, artists, or content creators who will benefit from this but Big AI companies and Hollywood...
@samleecole An initial scrape of data for Midjourney and OpenAI included private and deleted posts, things Tumblr doesn't have rights to. Unclear if this was turned over or was caught, with cleaning work ongoing [image]