The Key to AI Software Security Isn't Speed, It's Maturity
SecurityWeek, Tuesday, October 20th, 2026
Secure Code Warrior's Matias Madou maps an eight-stage AI coding adoption model to the risk levels and developer skills each stage demands.
Matias Madou says agentic AI development is expanding the attack surface faster than traditional controls, citing Gartner's forecast of a fourfold rise in agentic AI incident costs through 2027 and SonarSource findings that top coding models produce severe flaws.
Governance helps, but organizations often do not know their stage of AI adoption or what it requires.
He outlines a three-phase, eight-stage model (AI Assisted, AI Native, Agentic) from minimal AI use to autonomous multi-agent orchestration.
Each stage carries rising risk and needs matching controls, such as early governance policies, verifiable secure coding and review skills, detection of developer over-trust, MCP activity tracking and policy enforcement for agents.