Epistemic Fences - Theory & Practice

In What is an Epistemic Firewall? (Aug 2025), Pixie described the concept as a “[porous/selective] boundary in the flow of knowledge [at the end of the pipe, on the receiver side]”¹.

An “epistemic fence” is the same thing (receiver side filter), but its boundary is sharp. Often a signal of something engineered or artificial. Today we’ll look at a couple of case studies, and discuss how we deal with them in general.

---

.1

The first case study is from Vidhvat. He wrote this post mortem a few weeks ago. It is about how he seeded a completely false narrative as a kind of marketing/epistemic experiment around the video game he was working on:

To test how easily factoids can be made to percolate into the substrate with little to no evidence backing it up, the stage was set for me to test my hypothesis. Words are magic spells after all, so saying “Plot twist: I was always a Zionist” despite the implied irony followed by a Matrix reference was not going to register as satire but I was in part compelled by my curiosity, so I went ahead with it.

(April 9, 2026) “In a Tryst with Geopolitical Misinformation Landscapes”

This is the kind of empirical test that I described in my grant proposal in 2025. In cybersecurity, we do “pen testing”, try to break into a system to test its security. Part of my proposal was “ethical epistemic testing”: make up something completely false about a good thing, see how many people believe it, then write up a report to tell them how they were duped, and help them patch the bug.

Vidhvat initiated this experiment (I just analyzed it with him after the fact) but neither he nor I expected how difficult it would be to reverse. It was fascinating to see how a simple joke tweet “I am a zionist - like city of zion in the Matrix movie!” mutated & extra details were made up as the narrative spread (“did you hear the game dev, AND the publisher send all their profits to the Israeli military?”)². It’s still going months later.

This one has a sharp boundary because it splits so cleanly. People either:

1. Don’t know this claim about Vidvhat, and don’t care

2. Do care, and are stuck inside the fence

3. Do care, and have confirmed the answer in 30 seconds

---

.2

Our second case study today is “Brent Dill”. This is an epistemic fence for an online subculture that I have been studying called “tpot”. It has the same structure of Vidvhat’s documented case above does³.

Anyone who works around OMI (the open source marketing lab, a category of companies/communities who study these patterns) can verify this for themselves. When you talk to someone “inside the fence”, the pattern is as follows:

Them: “why are you protecting this evil person who sexually assaulted Octavia??”
You: “wait, how do you know this?”
Them: “everyone is talking about it! Everyone knows. Are you one of those people that doesn’t believe the victims? ”

You go and talk to Octavia, she says

“lol what. Who’s saying that? That’s crazy. This story keeps getting more ridiculous. I keep hearing new details all the time now”

You back to them and they say

“oh, huh. Well, even if it’s not true, I think there’s a lot of other reasons why he’s not good to talk to. It’s not worth the trouble”

This is how the “periphery” responds.

Those closer to it, those who have any access to facts (like the “Octavia” in this case, who I’m using as a fake name because I haven’t gotten permission to quote her here), say:

“yeah..I confirm this thing isn’t true. And I have seen many elements of this thing be completely misrepresented or flat out fake. I don’t know why this is happening. I don’t want the trouble/attention though. I’m glad someone is looking into it though! You’re doing great work”

---

.3

How do we deal with an epistemic fence? We do not tear it down, nor poke at it, unless we are willing to take responsibility for why the fence is there.

In Vidvhat’s case, we were there at the creation of the fence, so we know why it exists.

I was not there when the Brent Dill fence was erected, and I have no communication channel with those who are upholding the fence⁴. As an open source social engineer, part of my hippocratic oath / safety protocol⁵ is that if I notice something is intentionally memoryholed, we do NOT unearth it, our of respect to the engineers who are maintaining it. It is not my territory, it is not my problem, nor my secret to disclose.

My problem is “Brent Dill” has fallen in my territory. Which means it interferes with my normal operations (my territory is maximally truth seeking & has a minimal gap between public perception & truth). I need to make a choice about whether to uphold the fence, or erode it.

If the Brent Dill fence needs to be maintained, I think there is a cleaner, safer way to maintain it/memoryhole things⁶. As an open source engineer, we ask that the dark community at least mark their territory so that we do not accidentally interfere⁷.

We already have one contract with one such company. They did not know how to disclose the range of their territory without “outing themselves” (without showing us what it looks like from the inside, revealing too many sensitive secrets) but we worked with them to do a “clean, one way information transfer”. To tell us where we should NOT look without leaking information about the thing they are hiding⁸. As long as they are handling it, we stay out of the way. They recognize our territory. We agree that if their secrets “leak” into our territory, that we will disclose it to them first, before going public⁹. We can even help them keep it quiet if we understand that they are containing damage from old mistakes, not making new mistakes.

These rules are just like a responsible bug bounty program at DEFCON.

---

.4

This is my report of the shape & properties of “Brent Dill” the meme. Proper handling of this case requires much greater expertise & resources than I have. I would like to hand it off to someone higher up the chain who can handle engineering the best/optimal outcome.

Until then, what you need to know about “Brent Dill” is:

• If you hear people talk about him as “dangerous”, he is just as dangerous to talk to as people like Yudkowsky, Aella, Michael Vassar, Scott Alexander, Sam Bankman Fried, etc

  • If you don’t know who these people are: great! Don’t worry about. They’re famous internet people from California. Associated with the “rationalists” with is either a beautiful network state thing, a wholesome open secret society, or a cult, depending on who you ask

• If you do intend to hire, or work with Brent, there will be a lot of noise. His background and track record are difficult to find because of this noise & epistemic fence shenanigans (his name is essentially used as a political chip by various different factions with competing motives)

  • It would be sort of like hiring someone from FTX / Sam Bankman Fried, if the lawsuit hadn’t happened. You know something happened in those rooms, but you don’t know who did what, and it’s difficult to find out without an investigative reporter looking seriously into it, or a few lawyers

---

.5

I think that’s it! Someone else on the periphery of my network is making a Youtube video/documentary about “The Legend of Brent Dill”, but they don’t have to / most people probably won’t see it. The fence is still safe¹⁰.

1. We came up with this to distinguish it from the basic:

(1) solid information gates (like: you don’t see what happens inside Apple’s confidential meetings, and you know you don’t know)

(2) source-filtered information (when Apple makes a press release, you know they are selecting what information to make public).

The third category is (3) receiver-filtered information - this is like, if Apple put out the necessary information, but you failed to interpret it because of a “bug” in your own epistemic process. The block is at “the end of the pipe” so to speak.

2. This is exactly what the NYT reported on in March 2025 - about “secret narrative engineering companies” that seed a story, communities take it on and run with it, without understanding its source. Often doing the dirty work for these companies without even realizing it. The technical term for that is performing the function of a “Useful Idiot” (wikipedia). In this case we know there was no “secret company” behind it, because we made the campaign. We witnessed the “natural mutation rate”.

See: “NYT is trying to explain psyops”

3. In general, it is easier to detect psyops when it’s happening to a different subculture than your own. This is why a critical milestone in the Human Memome Project is getting every subculture on earth to be in contact with “a control group” and they mutually monitor each other (but doing so in a way where they don’t get “absorbed” and become just one culture/information network). You have full visibility if you can see the inside view, and an outsider’s view on you. Each network sets up a lighthouse that can function as a mirror for other networks.

4. It is totally possible that the fence is not “upheld” by anyone! It’s just the natural mutation rate. If so, when you go and try to poke it, you will encounter “no agentic resistance”. One way to test for “agentic resistance” is to apply the “is there a string on this plastic bag” thought experiment / test.

5. This should be a document that would kickstart this entire industry “in the light”, so it can go out of stealth. We can all come to a “geneva convention” rules for narrative warfare, with the expectation that not everyone will follow it, but those who follow it will spot each other and support each other, and the remaining “dark” networks become more visible once the lighthouse network links up.

6. You might be surprised to hear me speak in favor of memoryholing. I wrote about this in Suppression of Truth is Necessary for the Best Possible World. TL;DR it’s cases like where you are the ship captain and you realize the ship is doomed, you don’t tell everyone (causes panic, seals the ship’s fate), you opt to communicate to those who may be capable of fixing it.

7. Basically: preregister your psyops with us. We won’t interfere, we promise.

Otherwise, as you can see, we will accidentally interfere, and neither side is going to be happy. We both lose here.

8. This is fairly technical, and relies on the A/B/U mechanism, which I am currently writing a book/user manual for. Let me know if you are interested in funding this, getting an early copy, or sponsoring it. I am setting this up to be the critical foundational piece for open, ethical systems of information control & maintenance that is a pre-req for all of my other projects.

9. The other company hasn’t taken us up on this offer, but it’s not just that you can pay us for inaction (look away from certain territories), we can actually HELP you maintain the fence! And we can do it cheaper! And more ethically!!! Like you don’t have to pay off people or threaten them. We can just say

“look, there’s bad stuff that way, corruption, whatever, it will hurt a lot of people if we look that way. They are aware, they are fixing it. The problem is shrinking, not growing now. They hear us, and we hear them. We have open lines now”

10. Update: this is now out. See “An Outsider’s Perspective on the Human Sacrifice Rituals of the Rationalist Cult”