Researchers have demonstrated that putting words in ASCII art can cause LLMs—GPT-3.5, GPT-4, Gemini, Claude, and Llama2—to ignore their safety instructions.
Research paper.
Sidebar photo of Bruce Schneier by Joe MacInnis.
Related Post
- November 23, 2024
Google Exposes GLASSBRIDGE: A Pro-China Influence Network of
Government agencies and non-governmental organizations in the United States have become the target of a nascent China state threat actor
- November 23, 2024
North Korean Hackers Steal $10M with AI-Driven Scams
The North Korea-linked threat actor known as Sapphire Sleet is estimated to have stolen more than $10 million worth of
- November 22, 2024
Friday Squid Blogging: Transcriptome Analysis of the Indian
Lots of details that are beyond me. Blog moderation policy.
- November 22, 2024
APT-K-47 Uses Hajj-Themed Lures to Deliver Advanced Asyncshell
The threat actor known as Mysterious Elephant has been observed using an advanced version of malware called Asyncshell. The attack