The report
Cisco Talos analyzed NIK C2, a command-and-control framework with conventional surveillance and host-control functions plus an AI request path. The Windows implant could build OpenAI Chat Completions requests from commands supplied by its controller and relay model responses back to the attacker. If the attacker did not provide an API key, the implant attempted to read OPENAI_API_KEY from the infected host's environment.
This design changes the identity of the API caller. A model request can appear under credentials associated with the victim machine even though its prompt originated with an unauthorized controller. It may also consume the victim's API quota. The AI component does not explain how the attacker first obtained host execution, nor does it establish that the model independently selected exploitation steps. The malware's other collection and coercion modules remain separate capabilities.
The Talos technical investigation supports the client behavior and observed infrastructure. The NIK C2 dossier distinguishes a built-in credential fallback from proof that a particular user was billed or that AI output caused a breach. A useful response correlates unexpected API requests with endpoint process, environment-variable access and C2 activity rather than classifying unusual token use as an isolated model problem.