Physicist's case: AI risk jumps orders of magnitude because information has no mass

AryHHAry · x · 2026-09-15

A theory-physics-trained author argues AI risk isn't a new kind of danger, but its intensity jumps orders of magnitude due to scale. Every previous dangerous technology had a physical bottleneck as a universal speed bump — nuclear weapons need expensive, visible uranium enrichment; bioterrorism needs BSL labs; chemical weapons need supply chains. Information, however, has no mass: the same model weights can be copied to a million GPUs within hours, and capability isn't conserved.

The post cites an open letter reportedly signed by OpenAI, Anthropic, Microsoft, AWS and 100+ companies stating organizations have only months to prepare for AI-driven swarm cyberattacks targeting hospitals, water treatment and infrastructure. It also mentions researcher Jacob Coxon's resignation from Anthropic, quoting that the people building AI are "genuinely convinced it could kill us all before the end of this decade," while executives deliberately soften the language in media.

Related event: Ex-OpenAI/Anthropic researcher resigns, warning of superintelligence gamble(11 posts)→

Original post →

More from AGI Musings

AGI Musings channel →