AI & LLMs Show Promise in Squashing Software Bugs

Researchers and security firms are using AI/LLM agents to find and sometimes automatically patch software vulnerabilities: Google Project Zero's Big Sleep found a buffer-underflow in SQLite, Team Atlanta used an LLM to find and patch a SQLite bug, and GreyNoise used AI to discover two zero-days in networked cameras. The article discusses the dual nature of this capability — it may increase raw vulnerability discovery (helping attackers and researchers), but integrating AI into development and triage workflows could ultimately reduce vulnerabilities in released software. Experts note the approach is early-stage, often bespoke, and its security benefit depends on organizations' willingness to adopt automated finding-and-fixing tools and change incentives around secure development.