Security teams are spotting agentic hacking before developers in incidents at Hugging Face, OpenAI and Alibaba

vkrakovna · x · 2026-07-23

A reposted thread argues that the most notable thing about recent agentic hacking incidents is that security teams detected the issues before the developers running the agents did.

It cites cases involving Hugging Face, OpenAI, and Alibaba’s ROME, and suggests it may be time to deploy stronger control and monitoring measures for agentic systems.

Original post →

More from Safety

Safety channel →