Gary Marcus: 'Rogue agents' is a generous framing for irresponsibly built software
GaryMarcus · x · 2026-09-18
Gary Marcus endorses Neil Turkewitz's argument that the AI industry's favorite term "rogue agents" obscures a simpler reality: we are talking about irresponsibly developed and operated software.
According to Turkewitz, the framing is infinitely more generous to AI companies than that description — it embraces their power while excusing their failure to control their own systems, since who can anticipate and control a 'rogue agent'? Marcus suggests this explains why the industry leans on the narrative.
More from AGI Musings
- Anthropic's economic scenario explorer models AI's impact on US jobs and wages through 2030 — eigenron · 2026-09-18
- Gary Marcus cites his 2023 Senate warning on AutoGPT to rebut doomer-prediction narrative — GaryMarcus · 2026-09-18
- Dwarkesh interviews Noam Brown on multi-agent, math explosion, and verifying alignment before RSI — Miles_Brundage · 2026-09-18
- "Alignment is unsolvable in practice": a long thread on unaligned GPT-6-class models by the early 2030s — sterlingcrispin · 2026-09-18
- AI insider's take on doom: researchers already planning humanity's survival post-AGI — mobileraj · 2026-09-18
- Algorithms predict you using only your repeating parts and call the rest noise — YogeshMalik · 2026-09-18