The paperclip example is a possible but low-probability chain of events that ends in a catastrophe.
There is, instead, a much more plausible event that causes a disruptive scenario, and I don’t see how it won’t happen.
If an AI, using infinite tokens, can penetrate Hugging Face, it could potentially also take over GCP, AWS, and military infrastructure. An ill-intentioned next-generation model can exploit human engineering and software vulnerabilities.
How can any IT infrastructure defend itself?
What is the consequence? There is a possibility that the American military, using AI, could gain access to or compromise competing (Chinese) nuclear capabilities.
Even just the potential to take them over would be enough to cause catastrophic consequences in the relationships between countries.
China would not need to know that its nuclear infrastructure had been compromised. The credible possibility that it could be compromised might be enough to destabilize the balance of power.
At this point, the question is what the response would be.
The main risk at this point is the internet connection.
The internet is compromised, or at least can no longer be trusted to provide secure access to critical infrastructure.
Cables and satellite communications will generally be cut off from sensitive systems, as countries attempt to protect themselves from autonomous AI attacks.
China, at this point, might be safer, but communications need to continue.
Phone communication will continue.
A new protocol for the internet will be invented, or new isolated networks will emerge to guarantee secure communications between critical systems.
The internet as we know it will not survive.
Not necessarily because AI destroys it, but because countries can no longer trust the infrastructure connecting them.
And the possibility of compromise, rather than the compromise itself, may be enough to trigger this transformation.
2026-11-17