AIs are Getting Better at Finding and Exploiting Internet Vulnerabilities

A blog excerpt highlights Anthropic’s finding that Claude Sonnet 4.5 can, in simulations, execute multistage network attacks and exfiltrate data (including an Equifax-like scenario) using standard tools by recognizing and exploiting public CVEs, underscoring the accelerating trend toward automated exploitation and the importance of timely patching and security fundamentals.