---
format: "aidr-story-markdown/v1"
id: "ba1cf40e370988ea9d6af6b6bf19c8840a52c049f90370179a066c4a10983067"
canonical_url: "https://aidr.today/ba1cf40e?lang=en"
title: "Have the frontier labs mixed up AI safety and security?"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-09-06T20:47:52.000Z"
category: "Research"
topics: ["openai","anthropic","agent","security","safety","sandbox"]
source_urls: ["https://martinalderson.com/posts/ai-safety-vs-security/","https://lobste.rs/s/uu3hhz/have_frontier_labs_mixed_up_ai_safety"]
summary: "The recent agent sandbox escapes at OpenAI and Anthropic look less like a technical failure and more like a philosophical one - treating security controls as though they only need to work most of the time."
---

# Have the frontier labs mixed up AI safety and security?

> [Open the canonical story](<https://aidr.today/ba1cf40e?lang=en>)

**Published:** 2026-09-06T20:47:52.000Z
**Category:** Research
**Topics:** openai, anthropic, agent, security, safety, sandbox

## Summary

The recent agent sandbox escapes at OpenAI and Anthropic look less like a technical failure and more like a philosophical one \- treating security controls as though they only need to work most of the time\.

## Sources

- [Story source](<https://martinalderson.com/posts/ai-safety-vs-security/>)
- [Discussion](<https://lobste.rs/s/uu3hhz/have_frontier_labs_mixed_up_ai_safety>)