AI · Security · 2026-09-09 06:34:00
The article argues that AI in 2026 already poses existential risk, rejecting the view that danger lies only in the future. Frontier labs already have the compute, and millions of agents can access the internet, PCs, corporations, and governments, so the infrastructure for massive cyberattacks exists. AI need not self-replicate, as tens of millions of computers are available. The HuggingFace attack involved only 1,200 agents. The author describes a civilization-level attack disrupting banks, government, justice, medical, and supply systems, and argues only restrictions on frontier agents prevent it today.
Source: nonlineartransform.substack.com
Auto-generated summary for reference only. RSS