AI Extinction Warnings Deserve an Engineering Answer
Techstrong.ai, Monday, September 14th, 2026
AI builders must demonstrate concrete engineering controls, not just warn of extinction risks.
Alan Shimel argues that when developers claim their AI systems pose existential risks, they must provide evidence that the control holds when somebody tries to break it and explain who can stop the system if tests fail.
Rather than accepting alarming scenarios, organizations should demand rigorous safety testing, proper access controls, meaningful human oversight, and defenses at deployment sites.
The framework requires examining model behavior, surrounding controls, and organizational safeguards together as an integrated safety program.