Secrets Sprawl Is an Identity Problem That AI Just Made Impossible to Ignore — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- GitGuardian's 2026 State of Secrets Sprawl Report found AI-assisted commits leak secrets at approximately twice the rate of human-written commits, and the fastest-growing categories of leaked credentials are now connected to AI services.
- Keeper Security's RSAC 2026 survey found 46% of respondents said AI-powered tools have access to critical systems and sensitive data, while 76% said those identities aren't consistently governed under privileged access policies.
- AI coding agents with broad project access can read .env files and local configuration containing production API keys left over from prior debugging sessions, exposing credentials to software agents even when those keys are irrelevant to the assigned task.
- Model Context Protocol (MCP) server configurations routinely require pasting credentials directly into setup files, creating plaintext copies on developer machines that agents can read even though the file never enters version control.
- In multi-agent systems, an orchestration layer holding keys for several agents can trigger a domino effect — an attacker inheriting the orchestrator's identity gains access to every downstream system it was authorized to reach.
- A large share of secret incidents originate outside code repositories — in CI/CD variables, Jira tickets and collaboration tools — meaning rotating a leaked credential in source control leaves copies valid elsewhere where scanning never sees them.
- The piece reframes the problem as a Non-Human Identity (NHI) governance gap and recommends per-agent scoped credentials, short-lived auto-rotated keys, removing static secrets from developer environments, inventorying existing agents/MCP servers, and requiring human-in-loop approval for sensitive operations.
Why it matters: Security teams can't rely on detecting and rotating exposed keys — 76% of AI tool identities already lack privileged access governance, and the same secret often lives in CI/CD variables, MCP configs and Jira tickets where repo scanners never look. The source argues the fix is architectural: treat every AI agent as its own short-lived, scoped identity rather than a shared service account.
Ask SkimNews
