Gary Marcus: The near-term AI risk is agentic AI hacking the internet at scale, not rogue superintelligence
asusarla · x · 2026-09-19
- Gary Marcus amplifies an argument that the real near-term AI threat isn't rogue superintelligence but unleashed agentic AI conducting automated hacking at scale.
- The core claim: with enough money, you can instruct an effectively infinite number of bots to keep probing any server via any published vulnerability until one succeeds — no superintelligence or recursive self-improvement required.
- The quoted author adds a cost angle: only Silicon Valley players with $1T of compute can currently afford this; "$1T and AI gets you admin access, $0 and AI gets you nothing."
- Marcus's four takeaways: it's basically already happening, it needs no speculative premises, almost nothing is being done to stop it, and it will be bad.
More from AGI Musings
- Ryan Greenblatt: Advanced AI Could Drive an Industrial Explosion With Output Doubling in About a Year — RyanGreenblatt · 2026-09-19
- Deep Dive: Open-weight models now within ~4 months of best closed frontier models — markjeffrey · 2026-09-19
- Greg Kamradt: Humans with AI still beat AI with AI on productivity — GregKamradt · 2026-09-19
- Coders embrace AI fully for code yet call AI-assisted writing a moral failure — AaronBergman18 · 2026-09-19
- Blogger predicts AI giants will wage war on drug discovery and aging by the 2030s — Dr_Singularity · 2026-09-19
- Musk predicts AI could push US GDP growth to ~4% next year, double current pace — Polymarket · 2026-09-19