Fracturing Software Security With Frontier AI Models

Unit 42 (Palo Alto Networks) reports that frontier AI models now demonstrate autonomous reasoning capable of full-spectrum security research, not just coding assistance. Key findings include: frontier models excel at finding vulnerabilities in open source code but show only marginal gains against compiled binaries; OSS faces heightened near-term risk due to source code transparency; and AI enables faster zero-day discovery, N-day exploitation, and complex exploit chain construction. The post walks through an AI-augmented attack scenario covering reconnaissance through data exfiltration using MCP servers and LLM agents. Recommended defenses include assumed-breach posture, SBOM adoption, collapsing patching windows, automated incident response, and shifting to memory-safe languages.