Axios (April 26) Some large language models already have the ability to exploit known security vulnerabilities. Computer scientists at U of I found that GPT-4 can write malicious scripts to exploit known vulnerabilities using publicly available data.
