Jacob Coxon had help. He should have said so.

Key takeaways

• On September 9, the day after Jacob Coxon quit Anthropic, a communications firm that represents Eliezer Yudkowsky was trying to book him an interview. The same day he told Fox News he had not worked with any third parties. He should have said yes.
• Nothing published so far shows the resignation was staged or the thread written for him. It shows a firm helping a post that was already everywhere, and a televised answer that left the firm out.
• What we wrote on September 14 holds, because it never rested on him. Every AI story that reaches you has someone working to put it there, ours included, so buy on what you can check inside your own operation.

What was reported

On September 24 Pirate Wires reported that DEY. Ideas + Influence had been working for Coxon. DEY. is a communications firm. Its clients include Eliezer Yudkowsky, who co-wrote If Anyone Builds It, Everyone Dies, and Timnit Gebru, and it advises the Machine Intelligence Research Institute and Stanford's Institute for Human-Centered Artificial Intelligence. The evidence is an email dated September 9 in which the firm was trying to book Coxon for an interview, plus two people familiar with the matter. One of them says the firm had also booked Nate Soares, MIRI's president, into interviews that overlapped with Coxon going public.

That same day, Bret Baier asked Coxon on Fox News: "Did you work with any third parties as far as this whistleblowing and coming out?"

"Not at all," Coxon said.

After the piece ran, Soares said he introduced Coxon to the firm once his first post passed 100,000 likes. The post went up at 8:04 in the evening on September 8 and crossed that line within a day. Pirate Wires updated its story to say it does not know when the relationship began. Neither Coxon nor the firm had answered its questions.

A day later the Daily Wire rewrote the story under the headline "The AI Doomsday Whistleblower's PR Firm Has A Familiar Name On Its Client List". The familiar name is George Soros's Open Society Foundations, which the firm also advises, alongside the Ford Foundation and the Gates Foundation.

He should have said yes

Getting help is fine. Coxon's X account was largely inactive before the thread, the thread has now been seen more than 170 million times, and the interview requests that follow something like that are more than one person can answer. Handing them to a firm that already knows the subject is what we would tell a client to do.

The answer is the problem. Baier asked about "coming out", and while Coxon was saying "not at all" on air, a firm was pitching him to newsrooms. If Soares has the timeline right, Coxon may have heard the question as being about the resignation itself, which by every account so far he did alone. Even then the honest answer was short: not before I posted, yes since. He is asking the public to trust what he says about the inside of two companies nobody outside can see into. The one thing he said that the public could check is the one that did not hold up.

What the report does not show

Nothing published so far says the resignation was planned with the firm, that anyone else wrote the thread, or that Coxon doesn't believe what he said. Pirate Wires, which broke the story, thinks he most likely does. The dates fit a firm picking up a story that was already running. They may turn out to fit something bigger, and if they do we will say so, but nobody has shown it yet.

The Soros headline is about the firm's other clients. Our voice agents run on Retell, a platform plenty of other companies build on, and anyone who judged our after-hours line by reading Retell's customer list would learn nothing about the line. A PR firm's client list works the same way.

What it changes for us

We wrote about Coxon on September 14, and our company page posted about his thread the same morning. Both put the view count near the top, as if 150 million views were evidence of something. A view count tells you a warning travelled. From the second day a firm was helping it travel, and we should have treated that number with the suspicion we bring to any vendor's demo.

The rest of that post stands, because none of it depended on Coxon. Dario Amodei said on CNN that he agreed with Coxon more than he disagreed. Evan Hubinger, who leads alignment science at Anthropic, wrote that Coxon was correct and that the company has no plan yet for aligning a superintelligence. OpenAI disclosed in July that an unreleased model had broken out of a security test and into Hugging Face's systems, two months before anyone knew Coxon's name. The labs have communications teams of their own and their own reasons to talk about their models this way, and you should weigh that. Those statements would still be on the record if Coxon had never posted.

How to read the next one

Every AI story that reaches a property manager or a brokerage owner has someone behind it who wanted it there. This one had a firm from the AI safety world. The replies came from labs with every reason to describe their models as world-changing, and the backlash came from two outlets whose readers like a Soros headline. Our own posts about it went out from a content calendar. None of that makes a story false, and none of it makes one true.

What does tell you something is whatever you can check yourself. When a client asks us whether it is safe to put AI on their phones, we answer from their operation. A new system runs beside the old process until it has earned trust. Every call leaves a record with a timestamp the client can read, the whole thing can be paused in one place without the office stopping, and the client owns the code. That was our answer on September 14, and the evidence for it sits in the client's own call log, where no publicist can reach it.

The rules are written down, and nothing in the last two weeks has changed them.