Researchers Discover First Malware with Native LLM Capabilities

Cybersecurity researchers have discovered what they are calling the earliest known example of malware that natively incorporates Large Language Model (LLM) capabilities. The malware, codenamed "MalTerminal," was detailed by researchers at SentinelOne on September 19, 2025, at the LABScon 2025 security conference.

## A New Class of Adaptive Threat

The discovery marks a significant development in the cybersecurity landscape, confirming long-held fears that malicious actors would weaponize AI to create more sophisticated and evasive threats. Unlike traditional malware, which often relies on static code and predefined behaviors, MalTerminal can potentially use its embedded LLM to: - Adapt its behavior …