AI risk isn't superintelligence — it's millions of capable models plus weak cybersecurity
vishalmisra · x · 2026-10-22
vishalmisra restates his AI risk thesis: the danger isn't a superintelligence deciding to exterminate humanity, but millions of capable models compressing years of research into days, combined with imperfect cybersecurity. He argues this scenario has now arrived with the recent PHP incident and a Python supply chain attack. In a follow-up note, "This is not a proof that we're safe," he lists the real concerns: a million copies of highly intelligent models, automated labs, economic competition, weak cybersecurity, and humans' tendency to override sensible controls when money or national security is involved.
Related event: Wave of Security Incidents Fuels Warnings on AI-Powered Cyberattacks(3 posts)→
More from AGI Musings
- "We are here": cocktailpeanut posts Forster's The Machine Stops — cocktailpeanut · 2026-10-22
- 'If we lose the internet forever tonight': researchers joke—and stress—about living through the singularity — ZeroStateReflex · 2026-10-22
- Ashlee Vance: AI Agents Threaten India's Outsourcing Core as New Hard-Tech Startups Rise — kyliebytes · 2026-10-22
- If attackers and defenders both have ASI, what's the equilibrium? — flowersslop · 2026-10-22
- Anthropic researcher tells PhD students degrees are no longer safe bets; skeptics flag conflict of interest — 2C_ornot2C · 2026-10-22
- Every's writers on AI writing: judging by process, plus a classroom logging idea — lmoroney · 2026-10-22