Simon Willison • 12/10/2025

The Normalization of Deviance in AI

Read Original

This article discusses the 'Normalization of Deviance' concept, as applied to AI safety. It warns that as companies treat probabilistic and potentially adversarial LLM outputs as reliable without major incidents, they lower security standards and skip human oversight, creating a dangerous cultural bias that confuses a lack of attacks with robust security.

0 comments
#security #llm #prompt injection
The Normalization of Deviance in AI

Comments

No comments yet

Be the first to share your thoughts!

Browser Extension

Get instant access to AllDevBlogs from your browser

Top of the Week

1
React vs Browser APIs (Mental Model)
Jivbcoop • 3 votes
2
React, useEffect, and Stale Closures: How to Avoid Them with Dependencies and useRef
Jivbcoop • 2 votes
3
Building Type-Safe Compound Components
TkDodo Dominik Dorfmeister • 2 votes
4
Introducing RSC Explorer
Dan Abramov • 1 votes
5
The Pulse: Cloudflare’s latest outage proves dangers of global configuration changes (again)
The Pragmatic Engineer Gergely Orosz • 1 votes