Prof. Mollick: No ASI Can Build an Unbreakable AI Watermark
emollick · x · 2026-08-13
Wharton professor Ethan Mollick poses a critical AI security question: Could ASI build an AI watermarking tool so advanced that no ASI could bypass it?
His definitive answer is no. This highlights a fundamental vulnerability in the theoretical arms race of AI content detection, suggesting that purely technical solutions for watermarking may ultimately fail against advanced generative models.
More from AGI Musings
- Agent Fixed a Prod Bug Overnight, But Devs Hesitate to Merge — Specialist_Agent3599 · 2026-08-13
- Gary Marcus Slams GenAI Advocates: 'Making a Lot of Stuff Up' — GaryMarcus · 2026-08-13
- From the 1980s to Today: Why Multi-Agent Systems Will Actually Work This Time — kohjingyu · 2026-08-13
- AI Agents Replicate Corporate Hierarchy: The Future of Autonomous Capital — paraschopra · 2026-08-13
- Gary Marcus: AI Revenue Isn't Fake, But Profits Can't Justify the Massive CapEx — GaryMarcus · 2026-08-13
- New Essay Explores How Scale-Dependent Algorithms Shape AI Intelligence Explosion Forecasts — willmacaskill · 2026-08-13