Cyber Risks · The Hacker News

OpenAI Disrupts Russian, North Korean, and Chinese Hackers Misusing ChatGPT for Cyberattacks

Summary: OpenAI said it disrupted three activity clusters misusing ChatGPT for malware development: a Russian-speaking group that built and refined a remote access trojan and credential stealer, a North Korean cluster linked to an August 2025 Trellix-reported campaign against diplomatic missions in South Korea, and accounts overlapping with the Chinese group Proofpoint tracks as UNK_DropPitch. OpenAI said its models refused direct requests for malicious content, but the actors assembled building-block code for obfuscation, clipboard monitoring, and data exfiltration via a Telegram bot, and one Cambodia-based scam network asked the model to strip em-dashes from output to hide signs of AI generation.

Why it matters: The models refused the direct request and the actors got what they needed anyway by assembling benign-looking pieces, which means refusal at the prompt is not the control point. And a scam network editing out em-dashes shows actors are already managing the tell-tales your detection assumes.

Read the original at The Hacker News ↗

Topics
AI Security
Related