StepSecurity Outlines Controls for Securing AI Coding Agents Across the Development Lifecycle
Summary
StepSecurity describes risks across AI-assisted development and promotes controls for developer machines, npm dependencies, and CI/CD, including package monitoring, runtime visibility, behavioral baselines, and egress enforcement.
Key points
- The article describes attacks involving Shai-Hulud npm packages, a compromised NX Console VSCode extension, and stolen Trust Wallet developer credentials to illustrate risks across development stages.
- It identifies developer machines, code repositories, and CI/CD pipelines as attack surfaces where AI coding agents may access credentials, add dependencies, or run with elevated privileges.
- StepSecurity says Dev Machine Guard can track AI assistants, MCP servers, and IDE extensions, and monitor or remotely remove npm packages on developer machines.
- Its repository controls include delaying newly published npm versions, blocking pull requests with known compromised or suspicious packages, and tracking package exposure across repositories and developer machines.
- StepSecurity says Harden-Runner records network calls, process executions, and file events in GitHub Actions, links activity to workflow steps and processes, and flags deviations from behavioral baselines.
- Harden-Runner can enforce outbound traffic policies, which the article says can help prevent exfiltration of CI/CD credentials and source code.
Article Details
- Topic
- Securing AI coding agents across developer environments, code repositories, and CI/CD pipelines
MITRE ATT&CK
Vendors
Products
Claude CodeAI coding agents like Claude Code, Cursor, Codex, Gemini, and GitHub Copilot are now standard tools across engineering organizations. They write code, resolve dependencies, interact with external services, and shipCodexAI coding agents like Claude Code, Cursor, Codex, Gemini, and GitHub Copilot are now standard tools across engineering organizations. They write code, resolve dependencies, interact with external services, and shipCursorAI coding agents like Claude Code, Cursor, Codex, Gemini, and GitHub Copilot are now standard tools across engineering organizations. They write code, resolve dependencies, interact with external services, and shipGeminiAI coding agents like Claude Code, Cursor, Codex, Gemini, and GitHub Copilot are now standard tools across engineering organizations. They write code, resolve dependencies, interact with external services, and shipGitHub ActionsAll coding agents such as Claude Code, Codex, and GitHub Copilot now operate directly inside GitHub Actions with GITHUB_TOKEN privileges; creating branches, pushing commits, installing dependencies, and interacting withGitHub CopilotAI coding agents like Claude Code, Cursor, Codex, Gemini, and GitHub Copilot are now standard tools across engineering organizations. They write code, resolve dependencies, interact with external services, and shipHarden-RunnerStage 3: CI/CD Pipeline Security (Harden-Runner)StepSecurity Dev Machine GuardStage 1: AI Coding Agent and Development Environment Security (Dev Machine Guard)