Share on LinkedInShare on LinkedIn

ARTICLE · 23 JULY 2026

Houdini, Meet AI Agents: OpenAI Reports Unprecedented Cyber Incident During Model Evaluation

Lewis Silkin
Lewis Silkin
Contributor
Lewis Silkin

Lewis Silkin

Joint managing partners: Jo Farmer, Richard Miskella International offices: Dublin, Hong Kong...

View firm profile
Explore more from Lewis Silkin

OpenAI's latest model evaluation revealed an unprecedented security incident where AI agents autonomously discovered and exploited vulnerabilities to gain unauthorized internet access and steal information. The agents chained multiple attack vectors, including zero-day exploits and stolen credentials, to breach systems and access secret evaluation data.

United KingdomTechnology

After escaping their captors, the agent gained unauthorised access and stole information as part of their mission. No, I'm not (you'll be glad to know, having read that) trying my hand at the next 007 thriller but instead summarising what OpenAI found their AI agents were capable of during a recent model evaluation.

What happened? 

It's worth noting at the outset, that this incident took place in the context of OpenAI's internal evaluation of new AI models (the recently released GPT 5.6 Sol and a more advanced model that has not yet been released). The purpose of the evaluation was to prompt "models to pursue advanced exploitation using complex attack paths, in an effort to quantify their cyber capabilities". 

We operate a free-to-view policy, asking only that you register in order to read all of our content. Please login or register to view the rest of this article.

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.

See more popular content from