'Dark LLMs' Aid Petty Criminals, But Underwhelm Technically

The report examines Unit 42 and industry findings on underground LLMs like WormGPT 4 and KawaiiGPT, noting they can produce grammatically correct phishing emails, basic malicious code, and ransom notes that lower the bar for novice attackers, but are technically underwhelming (prone to hallucination, requiring human oversight) and currently show scant evidence of significant real-world impact—meaning existing detection and defensive measures largely remain effective.