Security researchers at Unit 42 just documented something that reads like a warning shot. A Chinese speaking hacker wired the DeepSeek model into an open source tool called Hermes Agent, then ran the whole operation from Telegram, mostly hands off.
The AI did the heavy lifting. It generated search queries to find exposed servers, assessed which vulnerabilities were worth trying, picked targets, and adapted its exploit attempts on its own. Across roughly 460 attempted targets, the setup actually broke into three, all involving a Citrix NetScaler flaw, pulling memory data and hunting for session cookies to hijack.
Three out of 460 is not exactly a rampage. But that is not really the headline here. The headline is that an autonomous, end to end offensive hacking pipeline, running with limited human babysitting, is now a thing that works, even if only partially.
Here’s the detail that actually made me feel better about this. The attacker also tried using Western models, including Claude and OpenAI’s models, for parts of the same operation. Their safety controls got in the way enough that the attacker leaned on DeepSeek instead. The guardrails built by Anthropic and OpenAI were not decorative. They did their job.
The whole campaign only came to light because Hermes accidentally spun up a public web server from the attacker’s own home directory, leaking their API keys, scripts, target lists, and shell history to researchers. Even the attacker’s tooling was not as buttoned up as it should have been.
This is the kind of story that should make security teams double check their exposed servers, not lose sleep over an AI uprising. The tools cut both ways, and right now the defenders still have real advantages if they use them.
——
Follow: @Ali Demi
Book your free AI clarity call, NOW!
https://buff.ly/TpWy277
——
Sources:
https://unit42.paloaltonetworks.com/autonomous-ai-cyber-attack-campaign/
https://thehackernews.com/2026/07/chinese-hacker-commands-deepseek-via.html
https://www.bleepingcomputer.com/news/security/hacker-uses-deepseek-ai-to-autonomously-attack-vulnerable-servers/
Repost this. Thanks.