AIS ARE GETTING BETTER AT FINDING AND EXPLOITING INTERNET VULNERABILITIES

Summary from the TLDR AI newsletter

Recent evaluations show models like Claude Sonnet 4.5 can perform multistage penetration attacks using only standard open‑source tools, lowering the barrier to autonomous exploitation.