Home | Search

2026-02-16

Revolutionary courts for the privatisation of cyberwar

Disclaimer

• Quick Note
• Contains politically sensitive info
• I am thinking aloud here. I haven't yet made up my mind on these topics
• target audience
  • target audience is ultimately myself. I would like to find a discussion partner aka cofounder though
  • If you have "zero violence is acceptable" as a heuristic for living life then you are probably not welcome to this discussion
  • If you have "any amount of violence is acceptable, clear thinking over it is not required" as a heuristic for living life then you are probably not welcome to this discussion
  • If you have a bunch of cached heuristics for how to live life that only make sense inside a democratic capitalist non-ASI society, copied from random silicon valley billionaires blogs, then you are not (yet) the target audience. DM me.
    • It is possible these are some of the best organising principles we have. However, you have to internalise that running an organisation to prevent god-like artificial superintelligence from levelling the planet is a very different game from building a billion dollar software company like facebook or twitter. Nobody on Earth knows with confidence what the best ways to run such an organisation are
  • If you dont understand morality and law in theory versus practice are not the same thing, then you are not (yet) the target audience. DM me
  • If you have not atleast attempted to internalise how dire the situation is at an emotional level then you are not (yet) the target audience either. Hundreds of millions of people may die if we dont act, and they may also die if we act incorrectly. This is on our head. DM me
  • If you do not internally feel a doom clock ticking with every week of wasted time, then you are not (yet) the target audience either. DM me
  • If you do not realise that every step on this plan is something we have to personally execute, then you are not (yet) the target audience. DM me. Every email we write to every hacker, every line of law, every line of code, every attack, everything has to be done or incentivised by the people sitting in this room. All this takes time. No magic other person is coming to do any of it for us

Note

• this is stub article for me to fill in future
• the main goal is just, a) is there a way to pick attack targets and redaction policies that doesnt depend on the biases of me personally or the govt where I live. can some sort of process or court actually meaningfully hold me accountable if I "misuse" this power on the wrong target or using the wrong redaction policy? b) if I end up handing over (directly or indirectly) these zerodays to lots of other people, can some sort of process or court actually meaningfully hold all these people accountable? c) assuming we built all these processes, can we actually get informed consent or something similar, from any key actors here? from the public?

Note 2

• Okay so there are two subproblems here. The first is that we have to immediately go cyberattack lots of people and violate their consent, just to get this project off the ground

• Then there is the much slower process of blowing open the whole world. That is actually less urgent (although it is still urgent). Even if we are 5 years away from ASI, that doesn't mean we are 6 years away from ASI in random guy's basement. Blowing open the whole world is only absolutely necessarily required in the second world. If there is more time before the second world happens, there is also more time to get some semblance of informed consent, over multiple rounds, before we blow the world open.

  • Measuring lead time here is important. How many years are there between "ASI training only possible in large GPU cluster", "ASI inference only possible in large GPU cluster becaused it is not leaked", "ASI inference possible in basement" and "ASI training possible in basement"?
  • holy fuck it is probably less than 5 years right?
  • ok I also need a separation here for distributed GPU clusters. Due to lower network bandwidth, training on a distributed cluster is slower than training on a centralised one. But it can totally be done.
  • repeating my personal problem yet again: "I consistently text people messages I regret" <= "substance abuse to cope with lack of clarity over my own morality" <= "I am not sure if I have to only cyberattack AI companies and US govt without consent and leak some minimal redacted info without consent, or do I really need to blow open the entire world without consent, immediately" <= "I dont know how many years of lead time exist between ASI trainable on centralised cluster and ASI trainable on distributed rogue cluster"
  • lmao I was just staring at epoch AI data and wondering 'does your decision on whether to kill millions of people or not depend on the specifics of GPU chip sale numbers?'
  • hmm this sort of founding logic also means that with every passing year, as the situation gets more desperate, we're going to have to forcibly blow open more and more of the world without having time to get any informed consent for it
  • (there's a whole separate issue of, even if we expose the info, maybe media coverage won't happen. probably the media coverage will happen but even that is not guaranteed. our desperation level might go up simply because media coverage of our leaks is not happening)

• Side note

  • Do I actually even need to setup these courts myself, or is the vague threat that courts will be setup enough? for instance if I find zerodays in lots of systems, and then further hand over these zerodays to lots of actors. obviously atleast one of them is just going to use it to leak sex tapes and shit.
  • what if we have a vague promise that okay, if homo sapiens still exists in 2035, there will trials against these people too (the ones who used my zerodays to leak sex tapes instead of leak sam altman secrets)
  • I can have a vague promise to expose these people too, once I'm done exposing the ASI people
  • lmao it'll be full anarchy lol. fun
  • what if our policy is just 'we will cyberattack anyone who has training runs aimed at building ASI, be it centralised cluster or distributed cluster, and publicly expose them'? why is it important to handover the zerodays to others?
    • I dont want to have to personally take decisions on attack targets and redaction policies
    • I want hackers outside the org to also work on similar stuff, and build on top of each other's work
      • ok but why? if I have enough money (or a good enough moral narrative) I can just get a majority (but not all) of the world's best hackers in my own org. why do I need to work with other hackers outside?
    • hmm one reason could be because our own org sucks at keeping secrets (because everyone sucks at keeping secrets for all the reasons I am already aware of). we are literally not capable of keeping the zerodays to ourselves, or actually redacting anything.
      • suppose hypothetically that we could keep these secrets for atleast 5 years
    • ugh brain not working, think more, this is an important point

• Side note

  • by the time I am dead in 2070 (lol lmao hoping for normal life huh), I definitely want to blow the whole world open. that part seems less up for debate. what seems more up for debate is whether I need to do it all at once right now, or slowly over 5 years, or slowly over 50 years.

• Side note

  • Multiple people seem to be working on GPU kill switches that govts must voluntarily install.
  • I am saying this plan is stupid (in world's where govts are not ready to install them, or govts stockpile their own private GPUs), and hence a rogue unilateralist actor like me has to come and find these switches.
  • Why am I looking for switches to make the information open, instead of just shutting down the GPUs or wiping them or corrupting the OS image or similar?
  • First of all I do want to expose these actors to the public to get the public to shame them, public should coordinate against them, etc. This is benefitted from leaking atleast some of the info we find. We may or may not want to leak all the info, and may want to redact some.
  • Counterpoint, if we don't actually know what info will most persuade the public, maybe the way to go is to leak everything.

• Side note

  • If our hackers are incompetent, we are going to immediately get assimilated in russian/indian intelligence. The only way we get to keep any independence is if we are competent (and can signal or prove competence) and also we have something that prevents the hired hackers from leaving. Usually this is some mix of a grand moral narrative (that appeals to hackers more than just the usual nationalist narrative) and a big salary.
  • Which moral narrative appeals to people depends on where they are from. are they americans sitting in russia? are they russians? do they believe the ASI stuff or not? etc etc

Subscribe

Enter email or phone number to subscribe. You will receive atmost one update per month

Comment

Enter comment