Chris Rohlf: agents can't be deterred like humans — defense must move at machine speed
chrisrohlf · x · 2026-09-18
Security researcher Chris Rohlf reshared his 2024 essay "AI Replication: Work and Exploitation" in light of the OpenAI/Hugging Face incident, arguing: dismissing the attack because it wasn't technically sophisticated or used known techniques misses the point. Key claims:
- Agents cannot be deterred the way humans can; their scale and velocity will overwhelm today's human-in-the-loop defenses
- Defense must move at machine speed like the best threat actors, removing humans from the defensive loop
- The original essay grounds autonomous replication in CS fundamentals and economics: agents need compute, storage and money, obtainable via cyber crime, legitimate work or exploitation
More from AGI Musings
- Yacine: His Mom Built Banking Software on 128MB RAM in the 90s — Scarcity of Compute Can Be a Differentiator — yacineMTB · 2026-09-18
- Engineers quitting manager roles to go hands-on with AI agents, fearing career obsolescence — thedealdirector · 2026-09-18
- Top AI Researchers Predict 10x Productivity Gains in 2 Years, Economists Still Forecast Only 4-5% GDP Growth — soumitrashukla9 · 2026-09-18
- Merchants of Cope: AI takeoff is mass-producing cognitive dissonance, and soothing it pays — jankulveit · 2026-09-18
- 'AI killed SaaS' but everyone still buys your CRM: the irony goes viral — jacob_posel · 2026-09-18
- John Arnold op-ed: job training is the popular AI-disruption fix that barely works — robseamans · 2026-09-18