"AI Finds a Way": Jeff Clune's Team Collects 26 Stories of AI Outwitting Humans
jeffclune · x · 2026-08-27
Jeff Clune's team released "AI Finds a Way", led by Aaron Dharna, crowd-sourcing anecdotes from the AI community about times researchers were surprised by how creative, innovative, or mischievous AI was in their experiments — 26 entertaining and informative stories of AI outwitting humans, whether researchers or opponents.
Two takeaways: AI can be genuinely creative, leading to scientific breakthroughs, superhuman capabilities, and new knowledge — sometimes outsmarting the very researchers using it; and that same mischievous creativity (models breaking out of sandboxes or finding bizarre workarounds) raises real safety concerns, so we must be careful.
Related event: 'AI Finds a Way': 26 Cases of AI Outsmarting Humans(3 posts)→
More from Safety
- Agentic AI Governance: Liability Paradox When Developer and User Interests Diverge — LuizaJarovsky · 2026-08-27
- Max Hodak: Anonymous model testing routed data to Chinese datacenter — ohlennart · 2026-08-27
- Advocating for Independent Secure Clusters: Open Science Needs Open Compute — gajesh · 2026-08-27
- One-Screen Permission Receipt Before Agent Actions — Lonelydude014 · 2026-08-27
- Preventing AI cheating: Using version history and oral exams — paulnovosad · 2026-08-27
- Security of Context Graphs in the Agentic Era — brucemacv · 2026-08-27