MalTerminal

MalTerminal is an early proof-of-concept LLM-enabled malware prototype, described by SentinelLABS as potentially the earliest known malware to generate malicious payloads at runtime. It was presented at LABScon 2024 and later discussed in SentinelLABS research on hunting LLM-enabled malware. The malware uses OpenAI GPT-4 via a hardcoded API key and a deprecated chat completions endpoint, indicating the sample was likely written before early November 2023. Identified artifacts include a compiled Python2EXE sample named MalTerminal.exe and related Python scripts such as testAPI.py and TestMal2.py.

When executed, MalTerminal offers an operator a choice between generating ransomware functionality or a reverse shell. It sends requests to a GPT-4 endpoint and dynamically generates malicious code in memory, including encryption, exfiltration, or reverse-shell logic, without needing to write the generated payload to disk. This runtime code-generation model was highlighted as complicating traditional static signature-based detection.

The malware is characterized in the source material as experimental and Level 1 (Experimenting) under the AIM3 maturity model. It is described as an LLM-driven ransomware and remote-access-tool concept in which AI is primarily used for code generation in response to human prompts. SentinelLABS also cited it as an example of the broader 'prompts-as-code' pattern, where prompts and API keys are embedded directly into malware.

There is no high-confidence reporting in the provided content that MalTerminal was deployed in the wild, sold, or operationally used by a known threat actor. The content instead consistently frames it as a prototype or PoC. Related variants such as TestMal3.py and Defe.py were described as defensive tools under the name FalconShield that asked GPT to assess whether Python code was malicious.