Security | Threat Detection | Cyberattacks | DevSecOps | Compliance

I vibe coded a working app in 10 minutes without writing a single line of code. It ran fine. It also had a hardcoded API key, a database password sitting in a Model Context Protocol (MCP) config file, and an npm package I never checked. In episode one of Secure Vibe Coding, we look at what AI-assisted coding means for secrets security, using data from GitGuardian's State of Secrets Sprawl 2026 report.

Ransomware is still growing, but payouts are shrinking. Enterprises got better at backups, so attackers moved to where defenses are thinner: SaaS, identity, and stolen credentials. Infostealers do the collecting at scale, and AI is widening the gap. Credentials for AI services were the fastest growing category of leaked secrets in 2025, up 81% year over year with 1.27 million exposed (State of Secrets Sprawl 2026). Most defense-in-depth programs were not built with this attack chain in mind.

We found 15x more valid secrets on developer laptops than in code repositories. In this September edition of What's New in GitGuardian, Sr. Product Managers Léna Cuissard and Emmanuelle Franquelin walk through two updates: agents that triage public secret leaks for you, and Developer Endpoint Protection coming together as one package.

Modern enterprise security is increasingly being tasked with keeping agents from affecting critical data and infrastructure. In this video, Dwayne, principal developer at GitGuardian, introduces how GitGuardian AI hooks extend the power of Docker Sandbox isolation. Their micoVM architecture plus the power of ggshield mean anyone can get an agent working in a secure way with very little effort. Chapters.