– What’s New, Amplified, and Diminished
Large Language Models (LLMs) are often discussed as a disruptive force in AI, but from a security engineering perspective, a more pragmatic framing is useful: an LLM system is still an IT system—just one with a probabilistic core and a natural language interface.
This shift in interface and execution model fundamentally reshapes the threat landscape. Rather than introducing entirely foreign risks, LLMs reconfigure existing security problems, while also creating a small but critical set of new attack surfaces.


