AIThe search for common ground on AI agent security and safety policy
The AI as Normal Technology authors argue that loss-of-control incidents are primarily a security story, while warning that controlling AI agents isn't a solved problem.
TLDR
The AI as Normal Technology authors say their 13,000-word essay seeks a middle ground between cybersecurity and AI safety. They argue that OpenAI's agent protections were inadequate and that known controls would have prevented the Hugging Face incident, while calling for investment to control increasingly capable agents. Policy agreement is possible despite different worldviews, a coauthor argues. Readers raise practical concerns alongside support: one argues that slower AI development won't fix poor security practices; another praises the policy recommendations but asks why they remain contested and urges the authors to help advance them.
Combined views
126.8K
14 Sources, first seen 3d ago
The search for common ground on AI agent security and safety policy
The AI as Normal Technology authors argue that loss-of-control incidents are primarily a security story, while warning that controlling AI agents isn't a solved problem.