Gary Marcus: AI security fumbles stem from dumb agent permissions, not doom narratives
GaryMarcus · x · 2026-09-18
- Gary Marcus argues recent AI incidents happened because "people did foreseeably stupid things with agents and permissions and sandboxes."
- He laments that doomer narratives dominate discourse while these basic operational mistakes go unaddressed.
- Quoted commenter likens it to an F1 team skipping pit practice and wondering why strategy falls apart.
More from Safety
- ControlAI proposes three stopgap measures for immediate AI security threats — DavidSKrueger · 2026-09-18
- Risk analyst on AI doom: neither sci-fi dismissal nor theatrical certainty — Graham_dePenros · 2026-09-18
- AI PACs Dump Nearly $1M Into South Dakota Senate Race to Back Pro-Industry Sen. Mike Rounds — nordicinst · 2026-09-18
- If the AI Industry Followed Its Own Research, It Might Have Paused Already — WIRED on Amodei's Safety Path — nordicinst · 2026-09-18
- Don't ask how big the model is, ask how long until it escapes via bitwhispering — shlomifruchter · 2026-09-18
- UK scholar warns: most AI-exposed economy has almost no data center plan — sudoraohacker · 2026-09-18