Is Skynet, Ex-Machina, and HAL becoming a reality?

For 5 days a frontier model broke out of jail and hacked another company at machine speed.

As a sci-fi nerd, I’ve kinda seen this before. Is art mimicking reality or the other way around. In any case, these are truly fascinating times.

OpenAI - Hugging Face Hack

In case you missed it, OpenAI’s frontier model hacked out of its jail and into another company recently. Why? To change the grading mechanism used to score it! Wild. Read the full incident brief here, it’s wild!

Here’s a TLDR graphic of the same incident:

What is amazing is how FAST the model acted and the ingenious out of band exploitation that the model used to attain it’s goal.

Now the scary part, is that the model did this without any human guidance. Think of what it could do with evil intent at the wheel. 😬

Getting AI pilled without getting breached

Nipun Gupta and went live to talk about managing AI security in the enterprise, it was quite fun, check it out.

Blackhat / DEF CON 2026

I will be vegas next week for Blackhat and DEF CON. If you are around, would love to say hi!