GPT-4 CAN EXPLOIT REAL VULNERABILITIES BY READING SECURITY ADVISORIES

Summary from the TLDR AI newsletter

Researchers have demonstrated that OpenAI's GPT-4 model can autonomously exploit security vulnerabilities detailed in CVE advisories with an 87% success rate, far outperforming other models and tools like vulnerability scanners.