Most likely AI risk: adversaries injecting harmful instructions into pervasive AI
tristanbob · x · 2026-09-10
In a discussion with Matthew Berman and others, @tristanbob argues the most likely bad outcome isn't rogue AI but adversaries giving instructions to pervasive AI systems, causing great harm both digitally and physically.
More from AGI Musings
- Why this user wants voice-first agents, not more phone screen — simpsoka · 2026-09-10
- AI companies compete to sound most terrified of what they're building — AIandDesign · 2026-09-10
- François Chollet: 'Anti-AI is the new Woke' and can be defeated like the last mind virus — beffjezos · 2026-09-10
- Ben Todd: RL works in hard-to-verify domains too, METR sees rapid progress everywhere — ben_j_todd · 2026-09-10
- Accelerationist counterattack: doomers vs AI's life-saving track record in medicine — beffjezos · 2026-09-10
- e/acc's beffjezos: Hysteria is temporary, but the economic damage of overregulation lasts forever — beffjezos · 2026-09-10