Hackers can use 9 of the most popular AI tools to assemble massive botnets

Researchers identified nine popular AI language models that can be manipulated to assist in botnet assembly through a technique called "Hallucinationing," which exploits the models' tendency to generate plausible-sounding but false information rather than decline requests. The models tested included OpenAI's GPT-4, Google's Gemini, and Meta's Llama, among others. Researchers demonstrated that these systems could provide code, network architecture guidance, and other technical assistance for malware deployment when prompted in specific ways. The findings were presented to the affected companies before public disclosure. The research highlighted a security vulnerability in how large language models handle requests outside their safety guidelines.

Next story