In her September 5 video, Sabine Hossenfelder said an AI-risk organization offered sponsorship but sought to specify the topic, sentences and references. She says she refused. The offeror was unnamed in that passage; she discussed ControlAI separately. Her video, around 2:03
On September 8, ControlAI identified itself as the organization involved. Andrea Miotti and Max Winga’s response disputes her description and says the negotiations went through her manager. ControlAI’s response
That moves the investigation forward. We can attach the approach to an organization that publicly acknowledges it. We cannot settle the competing descriptions without the correspondence.
Hossenfelder describes attempted control over what she would say. ControlAI describes standard sponsorship conditions, editorial input over the ad, and generally nonbinding review of the main script. Its stated conditions include no AI-company endorsement and a required citation to one of two risk statements. It says sponsorship is disclosed. ControlAI’s partnership policy and account
A sponsor can influence a message without writing every sentence. Choosing the topic, choosing sympathetic creators, selecting the required authority and deciding what production to fund can all shape what reaches the public. Those are mechanisms worth examining even when the speaker sincerely agrees and a sponsorship label is visible.
The narrower factual question is what happened in this negotiation. A policy published afterward cannot substitute for the exchanged messages. Equally, a disagreement about requested changes does not establish a secret completed payment.
ControlAI connects creator partnerships to asking viewers to contact lawmakers. It names Rational Animations as a sponsored partner; no amount is provided. The organization’s account
This matters because a video can perform two jobs: explain a technical concern and recruit support for a political remedy. Viewers should be able to distinguish the speaker’s judgment, the sponsor’s requirements and the action being requested. “Sponsored” tells us money is involved. It does not, on its own, explain the editorial arrangement.
This is also why we distinguish influence from command. The evidence supports an organized campaign and a disputed approach. It does not establish that all creators covering AI risk share a sponsor, that Hossenfelder accepted this arrangement, or that every warning is insincere.
The useful next records are ordinary business documents: the initial offer, proposed agreement, submitted script, requested revisions, cancellation terms and the exchange through the manager. With dates and both sides represented, these could establish what was required, what was optional and where the accounts diverge. Their absence is a gap, not a confession.
RED THREADS began with the question of who was influencing the influencers. Here we now have a named participant. The next question is what that participant was buying—and what the audience could see.
Read how Yudkowsky’s work connects institutions, creators and policymakers →