AI Gets Better by Default While Cybersecurity Decays — a Key Disanalogy for AI Safety
davidmanheim · x · 2026-10-14
Safety researcher David Manheim adds a crucial caveat to the analogy between today's AI and dotcom-era cybersecurity: by default, AI systems improve in generalizable ways over time even without effort, while cybersecurity gets worse over time without sustained investment.
- The original argument holds that dotcom-era security research — tools, monitoring, impact analysis — shaped how the cybersecurity industry developed, offering lessons for AI
- Manheim stresses this disanalogy matters: the two systems' unmaintained trajectories run in opposite directions, so lessons can't transfer directly
Related event: AIES Keynote Draws Lessons for AI Safety from Internet Security History(2 posts)→
More from AGI Musings
- Mathematicians lack AI literacy, argues researcher amid debate on AI upending math — gerardsans · 2026-10-14
- Google DeepMind team: AI agents boost ML engineering but can't run autonomous R&D end-to-end — HaydnBelfield · 2026-10-14
- DeepMind survey of 12 frontier researchers: execution automates first, ideation stays human — HaydnBelfield · 2026-10-14
- Artificial neurons don't physically exist — an LLM is just matrix math — ValerioCapraro · 2026-10-14
- First clear sighting of a rogue AI agent fleet from China scraping Amap on Tencent Cloud — emollick · 2026-10-14
- Heavy ChatGPT user since 2022 asks: has AI been a net positive for your life? — Alternative-Car8221 · 2026-10-14