China-based hacker employs DeepSeek in autonomous threat campaign

A Chinese-speaking threat actor has been identified running an AI-based hacking campaign that leveraged DeepSeek, using the Hermes Agent framework for autonomous cyber capabilities, according to a report released Thursday from Palo Alto Networks’ Unit 42.

The hacker used Hermes Agent to search for critical vulnerabilities, scanning GitHub for recently trending proofs of concept, according to the report.

The hacker attempted to leverage Claude Code and Codex in a limited capacity, demonstrating an effort to explore Western AI tooling. Claude Code was used to test connectivity and for proxy validation, according to researchers.

The threat actor used China’s DeepSeek AI platform to exploit a Langflow vulnerability tracked as CVE-2026-33017. The attempt ultimately failed, as the vulnerability required auto login to be enabled or a public flow ID. The hacker had neither, according to researchers.

DeepSeek was then used to conduct autonomous research, looking for higher-value vulnerabilities. DeepSeek searched GitHub for trending proof of concept repositories before finding a POC for a chained vulnerability in n8n, which is an open-source workflow automation platform.

“We do not know if that targeting was intentional or simply a random list of targets that had internet-facing services the attacker was interested in,” said Andy Piazza, senior director of threat intelligence at Unit 42.

Critical flaws

The report provides a window into how threat actors are attempting to leverage AI-based tools to conduct attacks. Researchers said they believe the attacker was independent and not linked to any state-supported entity.

The chained attack sequence involved an arbitrary file read flaw, tracked as CVE-2026-21858, and a remote-code execution flaw, tracked as CVE-2026-68613. DeepSeek found potential targets, but the attacks ultimately were unsuccessful because authentication was required.

The hacker also conducted manual operations, with successful exploitation of a critical memory overread flaw in Citrix NetScaler, tracked as CVE-2026-3055.