The Pulse
NIK C2 Malware Routes OpenAI Prompts Through Infected PCs
Cisco Talos says NIK C2 is a Windows command-and-control framework that sends an operator’s prompts to OpenAI from an infected computer. If the operator supplies no API key, the malware can use a key stored in the victim’s OPENAI_API_KEY en
AI.info Team · Written with AI · fact-checked ·
NIK C2 can make an infected Windows computer send an operator’s prompts to OpenAI—and, if the operator provides no API key, use a key stored on the victim’s machine. Cisco Talos described the framework on October 29, saying it had found NIK C2 infrastructure active online and assessed with moderate confidence that the malware was a functioning tool, not just a proof of concept.
The distinction matters: Talos describes code designed to draw on a victim’s OpenAI account, not evidence that a particular victim was charged. The malware also does not hand control of the computer to an AI model. Instead, an operator directs prompts through the implant, which sends them to OpenAI and returns the generated response to the operator’s command panel.
NIK C2 sends prompts from the victim’s machine
Talos says the 64-bit Windows implant is written in Go and connects to its command-and-control server over encrypted WebSockets, with an HTTP polling fallback. Its AI feature makes a separate HTTP request to OpenAI’s Chat Completions endpoint from the compromised computer, then passes the model’s response back to the operator. The request runs synchronously, with a 60-second timeout.
The command payload can include a prompt, an API key and a model name. When no key is supplied, the implant checks the machine’s OPENAI_API_KEY environment variable. Talos says a key found there could fund the operator’s requests through the victim’s OpenAI account, while the provider’s access logs would show that key. The code also records token-usage data from each response and returns it to the command panel.
“Not merely a proof of concept”
“Observing the NIK C2 infrastructure active in the wild allowed us to assess with moderate confidence that NIK C2 is not merely a proof of concept, but a fully functional, AI-integrated C2 framework.”
Chetan Raghuprasad, Cisco Talos
Talos says the implant supports more than 52 commands, spanning data theft, surveillance, remote control, evasion and lateral movement. Its listed capabilities include stealing browser credentials and cryptocurrency wallet data, logging keystrokes, capturing screens, monitoring clipboard contents and routing an operator’s traffic through the infected machine.
The AI feature is one command among that larger set, rather than the framework’s sole or primary control mechanism. Talos says the operator can choose a model in the command payload; absent a choice, the implant’s embedded default is GPT-4. That design lets the operator change the requested model without replacing the malware, according to the analysis.
The ransom screen does not encrypt files
NIK C2 also includes a screen-locking ransom module. Talos says it displays a customized demand for Bitcoin or USDT and repeatedly relaunches the lock screen if someone closes it. The module does not encrypt files, so its ransom tactic is to deny access to the desktop rather than make stored data unreadable.
Talos’s findings point to two separate risks for defenders: the implant’s broad theft and monitoring functions, and the possibility that its AI requests could use credentials already present on a compromised computer. The company published the ClamAV signature Win.Trojan.GoDiscorder-10060526-0 and linked indicators of compromise from its research page.