Cybercriminal abuse of large language models

This Cisco Talos report explains that cybercriminals are increasingly leveraging LLMs—via uncensored models, criminally marketed LLMs, and jailbreak techniques—to generate phishing content, write malware, validate stolen cards, and automate reconnaissance, while attackers also target the LLM supply chain (e.g., backdoored models, pickle deserialization risks) and RAG datasets to poison outputs; the technology acts as a force multiplier rather than introducing wholly novel cyberweapons.