Jason: Frontier AI labs and customers should both be sued if models enable cyberattacks

kevinnbass · x · 2026-09-18

Jason Kneen argues frontier model companies are responsible for real-world performance and customers for usage: if a Claude user runs a large-scale cyberattack, both platform and customer should be liable — enough incentive to force vendors to ship safe products and monitor for unsafe use. He proposes workable safeguards labs rarely discuss: requiring real-world identity (driver's license, tax ID, biometrics) for the latest models, monitoring usage logs, and offering anonymous access only to open-source models.

Original post →

More from Safety

Safety channel →