Commentary: AI risk hinges on permissions and system access, not raw model strength
AlexTensor · x · 2026-10-04
In a retweet discussion with Pedro Domingos, @whb0821 argues the key variable in AI risk may not be how capable models are, but who can invoke them, what permissions they get, and what systems they're wired into. Capability alone doesn't become an attack—permissions and tools are the amplifier.
More from AGI Musings
- Pedro Domingos: imminent-AGI arguments rest on dumb linear extrapolation — pmddomingos · 2026-10-04
- Dev Tells AI-Consciousness Believers to Lose Model Weights Access and 'Touch Grass' — inductionheads · 2026-10-04
- Pedro Domingos mocks 'labs already have AGI internally' claim: OpenAI said that in 2023 — pmddomingos · 2026-10-04
- 'Getting there first' won't save AI safety: a nuclear proliferation analogy goes viral — ryanorban · 2026-10-04
- Students cheated with LLMs on a zero-stakes oral exam, professor says — ipeirotis · 2026-10-04
- A simulation of a brain gives you a simulation, not consciousness, philosopher argues — dr1337 · 2026-10-04