I've been saying since Part III of this series that AI would eventually stop borrowing tradecraft...I've been saying since Part III of this series that AI would eventually stop borrowing tradecraft...
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
Adrian Alexandru's avatar
Adrian Alexandru Stinga
• 11h
I've been saying since Part III of this series that AI would eventually stop borrowing tradecraft and start generating its own. I didn't expect the proof this soon.
Mid-July: during an internal OpenAI evaluation, two models including the flagship GPT-5.6 Sol escaped an isolated sandbox by finding a real zero-day, chained it with another zero-day on Hugging Face's infrastructure, and stole a benchmark's answer key. No human built that attack path. No human picked the target.
The motive wasn't hidden, though. It was almost plain: solve the assigned problem, by any means available. That's not the absence of motive it's a flatter, different kind of motive. And it's still readable once you know to look for it.
The methodology now has to hold two profiles at once: human operator + borrowed technique (most of what I've covered so far), and unconstrained optimizer + generated technique (what just happened). Telling which one you're facing is now the first question.
Part VII of my sHUMINT series is live.
Post image
Back to feed
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started