Hacker uses DeepSeek AI to autonomously attack vulnerable servers

A Chinese-speaking threat actor is using the DeepSeek AI model together with the open-source Hermes Agent to run autonomous cyberattacks against exposed servers, according to BleepingComputer. The operation is described as requiring limited human involvement, meaning the AI stack is doing much of the work once a target is in scope rather than an operator driving every step by hand.

Technically, the stack pairs a general-purpose model, DeepSeek, with Hermes Agent, an open-source agent framework that can plan and execute multi-step actions. In this setup the model supplies reasoning and decision-making while the agent layer turns those decisions into concrete actions against reachable hosts. Autonomy here means the chain can probe, decide, and act on exposed servers with only light human oversight, not that every stage is fully unsupervised or that the tooling is novel in isolation.

For engineers and builders, the signal is that commodity model APIs and public agent tooling are already usable as offensive automation against anything left open on the network. Exposed management interfaces, unpatched services, and weak remote-access paths become more attractive when an agent can work through them at machine pace with minimal operator time. Defensive work has to assume that reconnaissance and exploitation workflows can be agentized, not only scripted by hand.

In market terms, this is the dual-use problem for open models and open agent frameworks made concrete: DeepSeek and Hermes Agent were not built as red-team products, yet they are usable as building blocks for low-touch attack campaigns. The cost and skill bar for sustained scanning and follow-on action drops when both the model and the agent are available off the shelf, which shifts pressure onto operators who still treat “obscure but exposed” infrastructure as low risk.

Practical takeaway: inventory and close exposed attack surface first—public services, admin panels, and remote management endpoints that an automated agent can reach without social engineering. Watch for further reporting on how Hermes Agent is being prompted or chained with DeepSeek in real campaigns, and for whether similar pairings of open models plus open agents show up against the same classes of targets.