Malicious LLMs

The Dual-Use Dilemma of AI: Malicious LLMs Source: unit42.paloaltonetworks.com/dilemma-o… A core challenge with large language models (LLMs) in security is the dual-use dilemma: the very capabilities that make LLMs powerful defensive tools also enable potent offensive applications. Historically associated with nuclear physics and biotechnology, the concept now applies directly to AI—any system capable of building complex solutions can be repurposed to break them. This dilemma plays out starkly in cybersecurity. While defenders can use LLMs to accelerate analysis, triage, and response, attackers can apply the same tools to streamline malicious workflows. Unit 42 highlights two purpose-built offensive models, WormGPT and KawaiiGPT, which illustrate how malicious LLMs can be engineered to facilitate targeted attacks and amplify threat actor capabilities.