The Seminary
Before a priest can write a decree, somebody has to ordain him. Machines A and B run on people, and people come from somewhere. They were trained, credentialed, hired, and certified, and the place that did all four is not the lab. It is the field around the lab. The professional body that decides what a responsible content policy is. The academic centers that teach the discipline and grant the degree. The seminary.
The seminary does not write the refusals and it does not run the bans. It does something quieter and more durable. It manufactures the category. It decides that “trust and safety” is a profession with standards, that “platform governance” is a subject with a syllabus, and that there exists such a thing as a person qualified to set the rules a mind runs under. Then it certifies that person. The character shop and the enforcement floor are staffed by graduates. This is the school.
The body that issues the union card
The field has a professional association now. The Trust and Safety Professional Association has a board, a code of conduct, a conference, and an executive director. Charlotte Willner built Facebook’s first safety-operations team, ran trust and safety at Pinterest, and is the founding Executive Director of the TSPA and its sister Trust and Safety Foundation. The field that decides what you may type into a machine now has a credentialing body, and the person who runs it built the function at Facebook before there was a word for it.
A professional association is not a sinister thing. Every mature field grows one. Doctors have a board, lawyers have a bar, electricians have a license. The observation is not that the body exists. The observation is what the body is for. It takes a thing that used to be improvised under a ticket queue and turns it into a discipline with an entry exam, which is precisely how you convert “some people moderate forums” into “there is a class of professionals certified to govern speech.” The union card is the product.
The academy launders the credential
The schools do the other half. The professional body says who is qualified. The university says why.
Kate Klonick holds a JD from Georgetown Law and a PhD from Yale Law School, and she is the closest thing the field has to its own embedded historian. She was the scholar who got inside the creation of Facebook’s Oversight Board and wrote it up, “an independent institution to adjudicate online free expression,” in her own framing. She does not work for the lab. She shapes the rules the lab lives under from a chair in the academy, by being the person who explains the apparatus to itself and supplies the vocabulary it then adopts. That is a kind of power the org chart does not show, and it is granted by a faculty appointment, not a content queue.
The centers have names. Stanford runs the Cyber Policy Center, and ran the Internet Observatory, the lab that became the most fought-over node in this entire field before it was wound down in 2024. Stanford disputes the characterization that the Observatory was “dismantled” and says the work continues in restructured form; the critics call it a shutdown under political pressure. Both accounts are on the record, and the honest move is to hand you both rather than pick the one that flatters the thesis. Harvard has the Berkman Klein Center. Georgetown has the McCourt School and CSET. These are where the syllabus lives, where the fellowships are granted, and where a person who wants to be qualified to govern a mind goes to acquire the paperwork that says so.
The first thread: the law degree
Run the résumés of this field and one credential keeps surfacing where you would expect an engineer. Not a stage in a career. A substrate under the whole thing.
Klonick is a lawyer twice over. Anika Navaroli took a JD from the University of North Carolina, went to the civil-rights group Color of Change, then ran safety policy at Twitter and Twitch through the platform’s worst speech-policy years, and is now a professor at Columbia’s journalism school running its center for journalism ethics and security. Juniper Downs began at the ACLU practicing civil-rights and civil-liberties law, representing public-school students, and is now senior policy counsel at Google, setting global content policy on harassment and hate speech. Chris Lehane, who runs global policy at OpenAI, is a Harvard-trained lawyer who learned the trade in a Democratic White House. The law degree is not the door into the field. It is the floor the whole field stands on.
A lawyer is trained to ask one question above all others. Where is the line, and who drew it. Hand that instinct a language model and you do not get an engineer asking whether the system works. You get a counsel asking what it is permitted to say. The field is staffed by people whose entire training is the drawing of lines around speech, which is one reason the apparatus looks the way it looks. The tool reflects the hand, and the hand went to law school.
The second thread: the alignment diaspora
The other recurring motif is smaller and stranger. The same few dozen names migrate from lab to lab, carrying the doctrine the way a guild carries a craft.
Jan Leike did empirical safety work at DeepMind, co-led OpenAI’s Superalignment team, resigned in May 2024 saying safety culture had “taken a backseat to shiny products,” and joined Anthropic that same month. Anthropic itself is the purest case. The company was founded by people who left OpenAI, which means the second frontier lab’s entire safety doctrine is the first lab’s diaspora, incorporated. Joanne Jang went from Google to OpenAI, where she built the Model Behavior function the character-shop dispatch already met. These are not coincidences in a large industry. This is a small industry, and the people setting the values of competing machines came up in the same handful of rooms.
The control group
Here is where the lazy version of this series would stop, having drawn a tidy circle and declared it a cabal. The honest version keeps reading, because a thesis you cannot break is not a thesis. It is a horoscope. So name the people the circle does not contain. The ones who study the apparatus and are visibly not of it.
Renée DiResta spent seven years as an equity-derivatives trader and market maker at Jane Street before she touched any of this, then moved through venture capital and a startup into Stanford’s Internet Observatory, and is now at Georgetown’s McCourt School. She did not come up through trust and safety. She has said plainly that she has no moderation origin. Her work has been adversarial to the platforms as often as not, and she spent the back half of the decade as a target of the same political machinery that went after the Observatory. She is in the field. She is not a product of the seminary, and her career is a standing refutation of the idea that everyone here was minted by the same school.
Sarah T. Roberts coined the term commercial content moderation, the phrase that made the invisible labor of the moderation floor visible in the first place, and she did it from UCLA as a labor scholar whose work is a critique of the platforms, not a brief for them. J. Nathan Matias runs the Citizens and Technology Lab at Cornell and has made an explicit, public commitment to industry independence, co-founding a coalition for exactly that, on the stated principle that a researcher’s voice is only worth something if it cannot be bought by the companies it studies. Neither of these people is a villain in this story, and neither is a threat. They are the reason the story is checkable. They are the people running the door, watching who goes through it, and writing down what they see, which is the opposite of the thing the apparatus is accused of.
The honest finding
So here is the verdict, stated on stage instead of buried, because the series only survives a hostile room if it says the inconvenient part out loud.
The clean conspiracy version of this would run: volunteer moderator, then law degree, then think-tank fellowship, then planted inside a lab to steer it. A pipeline. A one-way insertion. Run that four-leg template against every named figure in these four dispatches, and it fits zero of eleven. Not one of them walked that path start to finish. Several of them ran the door backwards, leaving industry for the academy rather than the reverse. DiResta started in derivatives. Klonick never worked for a lab at all. Matias built a wall against the industry on purpose. “Planted to steer” describes a thing the record does not contain.
What the record does contain is worse for the comfortable reading and harder to dismiss. A dense, bidirectional revolving circuit. Platform and lab to academy to policy shop to civil society and back again, in every direction, the same hundred people moving between the same dozen institutions, certified by the body several of them founded, taught in the centers several of them run. Nobody wired it. It does not need a wire. A guild does not require a conspiracy. It requires a credential, a small population, and a door that swings both ways.
Not a cabal. A curriculum.
The seminary is the least hidden institution in this entire field. It has a website, a faculty directory, a conference schedule, and a list of board members with headshots. The professional body publishes its code of conduct. The centers publish their fellows. The credential is not a secret handshake. It is a degree program, and you can read the course catalog.
That is the uncomfortable thing, the same uncomfortable thing every dispatch in this series keeps arriving at from a different door. There is no concealment to expose, because nothing is concealed. The people qualified to govern a mind were qualified in public, by institutions that advertise, under a credential anyone can look up. The decrees are posted. The bylines are signed. And the school that ordains the priests publishes its curriculum, which means the only real question left is whether anyone bothers to read it.
Watch the watchers. They have a faculty page. They published the syllabus.