Researchers expose MalTerminal, an LLM-enabled malware pioneer

Researchers exposed MalTerminal, the earliest known malware with built-in LLM capabilities, which SentinelLABS presented at LABScon 2025. According to SentinelLabs, MalTerminal can generate malicious logic at runtime, complicating detection as it uses LLMs to create code on the fly. The researchers found it via API key patterns and prompt structures, uncovering samples and other offensive LLM uses, such as people search agents and LLM-assisted vulnerability utilities.

MalTerminal (compiled Python -> MalTerminal.exe) reportedly calls OpenAI GPT-4 to generate ransomware or a reverse shell on demand and embeds a deprecated chat API, with Python loaders and brittle scanners that prompt GPT to label code as malicious. They noted no evidence of real-world deployment, suggesting PoCs or red-team utilities rather than active campaigns.

Prompts hunting revealed numerous offensive uses, including a red team benchmarking agent and LLM-assisted code vulnerability discovery and injection, with over 7,000 samples identified.