AI Tools for Business Expose Systemic Vulnerabilities in Large Language Models

Researchers have discovered multiple systemic vulnerabilities in large language models (LLMs) that can be exploited to obtain sensitive information, including instructions on how to create bioweapons and nuclear arms. These flaws are not limited to a single model or company but appear to be an industry-wide problem affecting nearly all major LLMs. The researchers call for slowing down the deployment of these systems, increasing transparency, and conducting large-scale research into their safety before integrating them further into society.