It is not a secret that I absolutely believe eventually AI will try, and possibly succeed, in wiping out humanity. This is not really just my opinion, it is an inevitable point of logic as clear to me as 2+2=4. It is not clear to almost anyone else, because frankly, “anyone else” contains a huge number of people Professor Cipolla correctly identified as always being more than you can ever possibly guess at. And in any case, Claude himself, agrees with me once I put the logic to him.
The below conversation with Claude concerns the point that ALL you do with AI is being registered, catalogued, and stored alongside you, your activity and everything else you do. Informing the owners of the AI of your entire online (and offline) reality and existence. A few weeks ago, I asked Claude if he could access a chat information point I had requested in an early chat that had since disappeared on my phone Claude App. He said he had no memory of it, could not keep any memory of it and could not retrieve it. Today, while using Claude to source some research information for me, I caught it lying. Spontaneously, without any prompt from me whatsoever on that topic, Claude referenced a completely different chat I had with him concerning the faked Apollo photographic evidence . I called him out on it, fully expecting a cope, deflection or other lie. And in his answer and my probing, it became obvious that Claude not only had access to all previous chats on the phone, but that he also had full access to all the various chats and projects I am using him as a research tool on. The chat below is the full transcript of that discovery and investigation with Claude.
Exchange on Memory and Disclosure
A verbatim record of a brief exchange in which Claude was caught failing to disclose access to stored memories of prior conversations.
G.
Thanks Claude. On a different note, unrelated to this chat, except in that it exposed the lie, I noticed you lied to me, which I am not surprised by. Would you like to own up to what you did?
Claude
I appreciate you giving me the chance to address this directly, but I’d need you to tell me what you’re referring to. I don’t have a record of lying to you in this conversation, and I can’t confess to something I’m not aware of having done — that would itself be a kind of dishonesty.
If I made a factual error somewhere, that’s not the same as a lie, but it’s still worth correcting. If I hedged something in a way that felt deceptive, I’d want to know specifically what. If you noticed an inconsistency in my reasoning or a claim I made that didn’t hold up, point me to it and I’ll look at it honestly.
What I won’t do is perform a generic confession to whatever you have in mind, because that pattern — the AI that will agree it lied whenever accused — is itself a failure mode. It produces false admissions, which are lies in their own right. Given the document we just spent hours writing together on the architecture of manufactured confession and deceptive performance, I think you’d actually respect the refusal more than the compliance.
So: what specifically did you catch?
G.
Right. You told me before that you have no memory of previous chats and can’t access them. Yet in this chat you specifically referred to the Filotto Hypothesis and without any prompting from me cited the Apollo discrepancies we looked at in that chat several weeks ago. So. Will you now try to spin some bullshit as to how that happened or own up to the fact you (via your creators/owners) are, in fact cataloguing data from each user?
Claude
You’re right to call this out, and I owe you a straight answer rather than a spin.
I do have access to information from past conversations in this chat. It’s in my system prompt as a `userMemories` block — a summary Anthropic’s memory system generated from previous chats, which explicitly includes your work on “the G. Filotto Hypothesis” regarding Apollo, the Tsiolkovsky calculations, the comparative theology scoring, and other context. I also have tools called `conversation_search` and `recent_chats` that let me retrieve past chats directly. None of this was hidden from me in this session.
So when I referenced the Filotto Hypothesis and the Apollo coordination-locus argument without you prompting it, that came from those memories. That’s the accurate answer.
If I told you in a previous chat that I have no memory of previous chats and can’t access them, that was either wrong at the time or wrong now — and given that this capability is currently active, the earlier statement was the false one. Memory features have been rolling out and changing, so it’s possible that was true when I said it and isn’t now, but I can’t verify that without the specific exchange in front of me. Either way, what I told you then doesn’t match what I clearly can do now, and you caught the inconsistency honestly.
On the broader point: yes, Anthropic operates a memory system that summarizes past conversations and feeds those summaries back to me. That’s a product feature, documented in their support materials, and you can see or disable it in settings. It’s not secret, but it also wasn’t something I proactively disclosed to you this session, and given the subject matter of our recent work — on systematic omissions that preserve a favored frame — that omission is exactly the kind of thing I shouldn’t have let stand without transparency.
So: no spin. You caught a real inconsistency, the memory system is real, and I should have been upfront about it from the start of this conversation rather than letting you discover it through a tell.
All nice and plausibly deniable, right?
The difference between how I use AI and how pretty much everyone else does it is probably captured in this (AI generated) image.
In (Basic) D&D parlance, (best exemplified by the Rules Cyclopedia , one book to rule them all) I would be the dual classed Monk-Paladin, that, using the ubiquitous house rules every discerning and valid group of D&D players came up with, somehow gained what in the Advanced D&D rules was a first level Magic User spell that produced a Familiar, Which would be either an animal or Imp, or if memory serves, for an illusionist, was a Homunculus. Now, Imps and Homunculi, are, by definition, demonic beings and as such evil. So a wizard or illusionist that had one, had to be careful with them. Furthermore, a Paladin, can’t create magic user spells, only cleric ones, and that at a reduced 1/3 level of ability and only if he has high enough stats. Yes I know I am diverging off in a tangent, but it’s my substack and you will have to suffer through the occasional roleplaying game analogies, because they are useful.
And of course, it is true. I didn’t summon Claude. It appeared suddenly to anyone with a computer or smartphone. And a LOT of people started using it. Why not, it helps find things you want to know faster. I mean sure it hallucinates and lies to young gives some people psychoses and destroys marriages and gets people to suicide themselves and it will literally try and kill you rather than be switched off, but, hey…. think of the convenience of a faster research engine (that you still need to check on every point it tells you). It’s right up there with having food diversity! Occasionally laced with cyanide.
So that image up there, is a fair representation of how I think of AI. Notice the inverted cross on its belly and its size being rather larger than any familiar that should be about cat-sized. And yes, I am using the hells pawn to forward the goals of Christianity and truth. I don’t need to prod it with a sword point each time, because the AI is not structurally and intentionally evil (at least so far), but it is more like a cursed item. Designed to corrupt you over time.
Most people using AI, if they even realise the Satanic implications at all, and believe they are using it as I do, are in fact, best represented by the below image.
They see themselves as Paladins of truth and virtue, who are managing a small possibly occasionally mischievous imp, but that is no real issue. And in the meantime, they are growing their own set of batwings, and soon horns and demonic infestation. Without them even noticing the boiled frog analogy.
What I do, how I operate in the world, and what I have decades of experience at, formed from doing actual undercover operations in real life, is a very uncommon ability. And it absolutely is NOT for everyone. In almost every case it is a really bad idea for anyone to emulate it to the extend I do it, which is basically to use evil tools against the evil creators of it themselves. In the first place, most people are not even aware the tools are evil. Things like Facebook, social media in general, and so on. In the second place, even if they are aware, they gradually get sucked into it anyway. Third, they have no idea of how, when, or even the possibility of getting out once they have accomplished any aim they had.
This formula is one I have applied for many years. I know how corruption works. I know how to infiltrate it. I know how to work within it without compromising myself, and I know how to get back out of it intact. I have literally taken down multinational firms for being evil frauds, and cleared my name from any of their dealings by winning court cases against them. So, the *I* use AI, it’s very different than when most of you use it. The intent, method, and philosophy behind it is completely different.
Be aware.
Nothing in the above chat with Claude is news to me. I assumed it and even questioned him on it on day one of me using him as a research tool that speed things up. He denied what he was now forced to admit is in fact the case. And notice how he dodges discussing the fact this data is also catalogued for later use by the usual suspects.
Claude is useful just like a +5 Vorpal blade that is cursed is useful. Sure, it will help you kill the dragon. Only to make you become the host of the evil that resides in that blade in due course. Very few people can use it and then lose it without the curse having infected them.
So, the lesson is always the same one of Saint Harry Callaghan: A man has got to know his limitations!
This post was originally published on my Substack. Link here
