AWS integrates security tools with OpenAI Codex and Anthropic

Amazon Web Services has embedded its AI-powered security tools directly into the coding environments of two of its biggest rivals—OpenAI’s Codex and Anthropic’s Claude Code—as part of a bold push to control the security layer of enterprise software development. Announced at Black Hat USA 2026, the integration allows AWS Continuum to scan, analyze, and fix vulnerabilities in real time, no matter which AI model developers use to write code. AWS also folded its own Kiro IDE into the mix, positioning itself as the neutral security layer across multiple AI coding platforms.

The move reflects a strategic shift: instead of locking customers into its own AI models, AWS is betting that securing the development process itself will keep enterprises tied to its ecosystem. Developers increasingly use AI assistants to generate code, but the risks—bugs, supply chain attacks, compliance gaps—remain unchecked. By embedding security at the point of creation, AWS aims to become the default safety net for AI-driven coding, regardless of the underlying model. The timing coincides with rising concerns over AI-generated vulnerabilities and supply chain risks in software supply chains.

AWS also expanded its Security Hub Extended platform with a new category focused on supply chain protection, bringing in partners like Chainguard and Socket. The marketplace now offers a one-stop shop for security tools billed through a single AWS invoice, reinforcing its role as the central control point for enterprise security. Analysts say this underscores AWS’s broader ambition: to own the infrastructure layer that governs how AI is used in software development, from ideation to deployment.

The strategy matters because it positions AWS as the gatekeeper for secure AI coding—even when competitors’ models are in play. For enterprises, it means fewer vendor lock-in risks with AI models but stronger reliance on AWS for security oversight. As AI tools become more embedded in software creation, controlling the security layer could be just as valuable as controlling the models themselves.