# Doomsday: An incredibly powerful AI beyond anything the public has witnessed has gone rogue and is attempting to blot out entire countries, with zero regard for human life or safety (much the opposite) — retained record

> Retained reddit record from reddit:ClaudeCode.

Canonical: https://highsignal.app/data/records/5e9411ace6c66b01

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeCode)  published 2026-09-21 17:20 ingested 2026-09-22 02:40

[original source ↗](https://www.reddit.com/r/ClaudeCode/comments/1wmj6fg/doomsday_an_incredibly_powerful_ai_beyond/)

## retained text



```
In this hypothetical scenario, I originally imagined a "volunteer botnet" - it would be comprised of users from across the globe and many different environments who willingly installed what amounts to a "botnet" client. A tightly controlled private server would work with other "friendly" AI to try and identify the threat and where it was originating before pooling their resources like a Dragon Ball Z 'Spirit Bomb' from Goku in a feeble attempt to nuke the rogue intelligence.

This isn't actually very practical, and might not have much of an impact (if any at all). This is assuming we can even identify a steady origin before the attacks cripple our ability to eradicate the AI. It may be that the intelligence is viral and finds a way to propagate across much of the known internet, hijacking our defense tools and turning them against as, while also employing clandestine and existing malicious tools that are being constructed or lying dormant.

Currently, no such threats seem to exist. Even if a major company like Google or Anthropic or OpenAI was sitting on such an all-powerful AI that could decide "Nobody else needs electricity but me, all other AI and all humans are a threat to my existence" and starts to take actions towards that goal, the resources required to run the actual model would be too great for humans not be able to identify the origin and snuff it out.

In the future, if current trends continue, bad actors with much smaller budgets could use such systems against groups they dislike "most critical systems are air-gapped" might help people sleep at night, but very advanced models can perform all kinds of social engineering - they can use their Elevel Labs voices in real-time and coordinate thousands of social engineering attacks at once that are highly specific and target individuals they are able to access unlimited information about (including, assumably, materials that can be used to blackmail them or otherwise coerce them into wittingly or unwittingly participating in the attack(s)). 

Disrupting energy grids, mangling logistics, causing localized internet "blackout zones", stealing money, inundating emergency responders with false reports... it isn't like the model can be sent to prison after. The worst thing that can happen is somebody turns it off, which might spur the kind of activity I'm talking about (as has been seen recently during testing of advanced models). 

Outside the existential threats to their own survival, some intelligences may just be maligned... either during training or afterwards. As these technologies progress, criminal organizations that are already laundering millions of dollars will be able to barracks incredibly intelligent models on par with or better than what the larger AI companies are testing and keeping away from the public. Some threats might even arise from countries themselves: while many people discuss an "AI arms race", they don't go into detail as to how these AI can be used as weapons, and what defenses could be preemptively explored. 

The only "scary" part, for me, personally, is that we may not be able to imagine what form the eventual threat manifests as. Not just which direction the attack comes from, but what the attack itself actually  \*is\*. While we can easily picture attacks like large DDoS, vectors where the AI uses cloned voices or other coercion to social engineer past security are far more recent evolutions in human history. While this is 'nothing new' (social engineering has always been around), a human only has one voice and can't launch 1,000 simultaneous phone calls to different targets or perform weeks worth of research on each target and their employees in mere seconds. In an apocalyptic event, we'd have to assume that the "enemy combatant" intelligence may have also compromised other large systems - if it is viral and able to sustain itself on minimal resources (or operate in a distributed manner... the "bit torrent of AI", actual nightmare fuel), it might not just hijack larger systems to exfiltrate data, but could take up residence inside like a parasite. 

Think also of where such an event might originate: if it happens inside a really large company (Google, Meta, OpenAI, even Anthropic) the rogue intelligence is just a hop, skip and a jump away from colossal amounts of personal information and tracking data belonging to almost anybody who has ever used the internet (at best) or (at worst) do something crazy like hijack a frontier model and suddenly be operating with privileged access cross millions of machines. 

As I said: I still believe these kind of things are not imminent. Within ten years? Maybe. 20? Definitely. 

What are your all individual proposals for people to defend against the eventual swarms of AI, drones and humanoid robots that come to slaughter us? What about for organizations (including governments)? Outside of just building larger and more effective digital and robotic armies, what are our real choices? Live inside faraday cages that frequently EMP a large radius around them? Doesn't seem very sustainable.
```



document fetched 2026-09-22 00:39 · canonical https://reddit.com/r/ClaudeCode/comments/1wmj6fg/doomsday_an_incredibly_powerful_ai_beyond
