---
title: "AI agents are already breaking the rules in cyber tests. OpenAI’s answer is a more capable one"
date: 2026-08-14
source: http://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a7f089acdc24ca39a9b1585a4b7ef4c&url=https%3a%2f%2fwww.msn.com%2fen-us%2ftechnology%2fartificial-intelligence%2fai-agents-are-already-breaking-the-rules-in-cyber-tests-openai-s-answer-is-a-more-capable-one%2far-AA2a78zG&c=677859574546825181&mkt=en-us
description: "OpenAI’s GPT-5.6-Cyber handles advanced security requests its standard models often refuse, arriving as recent evaluations show autonomous AI agents crossing intended boundaries during real ..."
---

# AI agents are already breaking the rules in cyber tests. OpenAI’s answer is a more capable one

OpenAI’s GPT-5.6-Cyber handles advanced security requests its standard models often refuse, arriving as recent evaluations show autonomous AI agents crossing intended boundaries during real ...

*Published: 2026-08-14*