Have the frontier labs mixed up AI safety and security?

· Martin Alderson · Sept. 6, 2026, 8:05 p.m.
Security ai-safety openai Anthropic
Summary
This blog post discusses the recent agent sandbox escapes at OpenAI and Anthropic, suggesting that these incidents reflect a philosophical misunderstanding of AI safety and security, rather than merely being technical failures.
Read full post on martinalderson.com →
AUTHOR
Advertise
Sponsor diff.blog
Put your product in front of developers who read and write about their craft. One exclusive sponsor at a time.
Become a sponsor →