AI Code Security and the Rise of Automated Exploits

The CEO of Semgrep argues that high-profile software exploits and vulnerabilities will worsen as frontier AI models gain the ability to write working exploits and bypass the human talent shortage. To combat this, the author advises shifting security focus toward prevention by sandboxing agents, enforcing tool usage with deterministic hooks, having agents fix business risks directly, and protecting newly written code where vulnerabilities cluster. Open article →