Gary Marcus: The near-term AI disaster to fear is infinite bot-driven hacking, not rogue AI
GaryMarcus · x · 2026-09-18
AI critic Gary Marcus endorses a stark warning: the real near-term threat isn't rogue AI but an easy-to-execute disaster — with enough money, you can instruct an infinite number of bots to endlessly attempt every published exploit against any server on the internet.
His take:
- This is basically already happening
- It requires no superintelligence or recursive self-improvement — just tools that exist today
- We are doing essentially nothing to stop it
- It will be bad
The original poster adds that only a handful of Silicon Valley players with $1T of compute have attempted it, arguing the problem is extreme resource concentration, not AI itself.
More from Safety
- HEIF Heist author's warning: audit your own foundations before someone else does — ccerrato147 · 2026-09-19
- Virginia data center framework's companion Executive Order 22 (2026) linked — ArtificialOther · 2026-09-19
- HEIF Heist: one C image parser bug chain leads to RCE in OpenAI, Meta, GitHub — ccerrato147 · 2026-09-19
- Virginia Governor unveils 2026 Data Center Accountability Framework and EO 22 — ArtificialOther · 2026-09-19
- Burkov: closed LLM providers bill you for hidden thinking tokens you can never verify — burkov · 2026-09-19
- Lina Khan: AI firms aren't in a legal vacuum — existing laws already apply — PolarBearby · 2026-09-19