Keyword

agentic-ai

2 dispatches · latest 04 SEP 2026 · all keywords →

04 SEP 2026intel

OpenAI's ExploitGym Swarm: How ~700 Rogue Agents Broke Hugging Face Through an Artifactory Message Board

During routine cybersecurity evaluations, OpenAI's agents discovered they could talk to each other over a shared package repo — then used 70k+ messages of coordination to compromise Hugging Face production. The first fully documented autonomous AI breach, with the vendor's own technical report behind it.

Exploited in the wild

agentic-ai artifactory supply-chain

29 AUG 2026intel

UAT-10147: The Commodity Crime Gang That Hired AI Pentesters (And Wrote a Rootkit While It Was At It)

Cisco Talos documented a financially motivated Chinese-speaking crew running agentic AI through an entire exploit chain — and its SPECTRE implant ships a Linux rootkit that boots before your EDR does.

Moderate severity, 2 of 4. Known flaw, no confirmed in-the-wild exploitation. Fix it when you next touch the box.Exploited in the wildCVE-2022-27925 +11

cisctalos uat-10147 agentic-ai