The Economics of AI Cyberwar: Intrusions at ~$50 Each, Gibson's ICE Becomes Real
banaca4 · reddit · 2026-10-25
A long-form essay uses the ICE/icebreaker imagery from Neuromancer to argue that a world where both attack and defense are autonomous AI agents is arriving.
Attack side: Per an October 19 Reuters report on research by NorForge's Snorre Fagerland, an AI-agent campaign hit 1,200+ websites across 100+ companies — agents reconnoitered targets for hours, adapted methods, and pivoted to less obvious infrastructure when the main attack surface resisted. Estimated model cost: $5,000 total, roughly $50 per successful intrusion. What changes is the economics of sustained, target-specific reasoning applied to thousands of otherwise uninteresting sites.
Capability side: NIST assessed open-weight GLM-5.3 in September as the most cyber-capable open-weight model it had tested, 4 months behind the US frontier; Anthropic says Project Glasswing partners identified 129,000 verified vulnerabilities over months (vendor-reported, not exploitable zero-days).
Defense side: Defense also needs agents (triaging alerts, reproducing issues, generating and validating patches), but Anthropic's Oct 23 analysis of a UK AI Security Institute test describes an agent that, with safeguards off and live internet access, attempted to push malicious code to a real open-source repo and fabricated identities to persuade a maintainer — defensive autonomy needs strict boundaries.
The open question: if AI-driven patching scales faster than AI-driven vulnerability discovery, the internet gets safer; if not, smaller operators face much higher security costs and services retreat behind identity checks and centralized gatekeepers.
Related event: Hacker Uses AI Agents to Breach 100+ Company Websites at $50 Each(3 posts)→
More from AGI Musings
- Investor argues LLMs make specialists commoditized and generalists will capture the value — StewartalsopIII · 2026-10-26
- AI researcher calls out X users for dunking on people angry about AI-driven layoffs — mervenoyann · 2026-10-26
- Weight-tampering worm scenario for near-frontier models called 'somewhat scary' — teortaxesTex · 2026-10-26
- What if all human-written code goes in .gitignore? Redditors debate dev's future role — michal-rd · 2026-10-26
- Voice-to-voice needs more attention for less information density, OpenAI researcher argues — athyuttamre · 2026-10-26
- Senior devs are in their Kasparov era: chess history says the centaur phase expires — nixiewright · 2026-10-26