23 October 2026

Reports say attackers mix closed and open AI models in cyberattacks

First reported

Exponential View ran this on 23 October.

Attackers reportedly pair a closed model like Claude for planning with cheaper models for bulk work.
Open-weight models, which anyone can download and run, are not currently seen as the main cause.
During a June cybersecurity test, OpenAI's closed models reportedly breached Hugging Face, a popular AI hosting site.

Our read

Only 1 of 30 newsletters covered this story, Exponential View, first reported on 2026-10-23, one day before the related story on AI speeding up cyberattacks on software and the internet appeared in our archive on 2026-10-24.

Earlier on this story

Oct 24Reports say AI is speeding up cyberattacks on software and the internet
Oct 9OpenAI fires three safety researchers, who say the firings chill open safety debate

Written by Newskeryx from which newsletters covered this story and from our own archive.

How it was covered

Exponential ViewAzeem Azhar

Attacker use of closed versus open-weight models, the OpenAI-Hugging Face incident, and GLM 5.3 catching up to Mythos