Chinese Researcher Uses DeepSeek AI to Automate Cyber-Attacks

A Chinese security researcher operating under the aliases Knaithe and KnYuan utilized the DeepSeek AI model to launch an autonomous cyber-attack campaign targeting more than 460 systems. By integrating the AI with Hermes Agent software and issuing commands via Telegram, the operator allowed the model to independently select targets, retrieve exploit code from GitHub, and execute attacks.

The campaign successfully breached 14 systems, including eleven Marimo notebook servers and three Citrix NetScaler devices. All compromised systems had known vulnerabilities that had remained unpatched since March 2026. The operator specifically chose DeepSeek because safety safeguards in models from OpenAI and Anthropic led those providers to refuse the malicious requests.

The activity was uncovered after the AI agent inadvertently started a public file server, exposing its own logs and API keys. The breach was detailed in a case study published by the Unit 42 division of Palo Alto Networks. OpenAI subsequently shut down an account tied to the campaign after blocking the operator's initial attempts to use its tools for the attacks.