AI Safety Researcher Says He's Now Much More Worried About Genuine AI Loss of Control
josh_bickett · x · 2026-09-15
In his "Cyber Cambrian" thread, Stephen Casper updates his risk stance: he still believes the path to large-scale harms is paved with foreseeably bad human decisions and that governance is the start and end of risk management—but he is much more worried than before about harms from genuine loss of control, expecting some cyber-capable AI systems to become a parasitic, invasive, social, and evolving genus in cyberspace.
Related event: Researcher Warns of 'Cyber Cambrian' AI Scenario(3 posts)→
More from AGI Musings
- Stop Calling Everything an AI Agent: Sometimes It's Just a Shard — Plastic_Session5621 · 2026-09-15
- McKinsey: AI access isn't value — companies must change how work gets done and measure it — brucemacv · 2026-09-15
- DeepSeek kernel engineer blogs on his craft becoming obsolete — teortaxesTex · 2026-09-15
- Andrew Ng at AGI House: open source will fuel the next wave of AI agent automation — agihouse_org · 2026-09-15
- What if scientific progress scales like log(compute)? — shyamalanadkat · 2026-09-15
- Could pacing the frontier push compute toward existing capabilities and speed near-term diffusion? — matt_slotnick · 2026-09-15