--- title: "Writing | Gazebo — AI Agent Security and IAM" description: "Guides and essays on AI agent credential management, MCP security, least privilege, and IAM for developers running AI tools." url: "https://gazebohq.com/writing" --- # Writing | Gazebo — AI Agent Security and IAM Guides and essays on AI agent credential management, MCP security, least privilege, and IAM for developers running AI tools. ## Related pages - [Writing Topics | Gazebo — AI Agent Security Guides](https://gazebohq.com/writing/topics.md): Browse AI agent security guides by topic — IDE tools and agent setup, secrets and credential management, IAM and access control, and the tooling landscape. - [IDE & Agent Tools | Gazebo Writing](https://gazebohq.com/writing/ide-agents.md): Secure credential access for Cursor, Claude Code, Windsurf, and other AI coding agents. Setup guides covering the MCP credential model and per-tool risks. - [Secrets & Credentials | Gazebo Writing](https://gazebohq.com/writing/secrets.md): How to store, scope, audit, and revoke API keys and credentials when your users are AI agents — covering the full credential lifecycle. - [IAM & Access Control | Gazebo Writing](https://gazebohq.com/writing/iam.md): Identity and access management for AI agents: least privilege, RBAC, zero trust, OAuth, and multi-agent pipelines. IAM concepts reapplied to autonomous agents. - [Tooling Landscape | Gazebo Writing](https://gazebohq.com/writing/tooling.md): Understanding which secrets management tools — HashiCorp Vault, AWS Secrets Manager, Doppler — cover which layer of the agent credential stack. - [Zero Trust for AI Agents: Practical Principles — Gazebo](https://gazebohq.com/writing/zero-trust-ai-agents.md): Zero trust for AI agents: every credential request authenticated, scoped to minimum, and logged per agent identity — never trust, always verify. - [Doppler and Gazebo for AI Agent Secrets — Gazebo](https://gazebohq.com/writing/doppler-vs-gazebo.md): Doppler handles secret storage and environment sync. Gazebo adds per-agent identity and audit logs on top. How to layer them without changing your setup. - [1Password and Gazebo for AI Agent Credentials — Gazebo](https://gazebohq.com/writing/1password-ai-agents.md): 1Password handles browser login credentials for agents. Gazebo handles API keys and tokens for coding agents. Different layers — here's how they work together. - [AI Agent Security Checklist: Cursor, Replit, and Copilot — Gazebo](https://gazebohq.com/writing/ai-agent-security-checklist.md): AI agent security checklist: per-agent identity, scoped credentials, access logging, approval gates on destructive actions, fast revocation. - [Why Environment Variables Are Insecure for AI Agents — Gazebo](https://gazebohq.com/writing/environment-variables-ai-agents-security.md): Env vars are a weak security boundary for AI agents — anything in the process environment is readable. A checklist for safer credential handling. - [AI Agent Secrets Management: 6 Best Practices — Gazebo](https://gazebohq.com/writing/secrets-management-best-practices-ai-agents.md): Six operational secrets-management practices for AI agents: prevent exposure, scope access, log activity, and revoke safely. - [IAM for AI Agents: Identity Architecture Explained — Gazebo](https://gazebohq.com/writing/iam-for-ai-agents.md): IAM for AI agents: the identity architecture behind scoped profiles, brokered credentials, approval gates, and agent lifecycle controls. - [Secrets Management for AI Agents: Core Controls — Gazebo](https://gazebohq.com/writing/secrets-management-ai-agents.md): Secrets management architecture for AI agents: vault storage, brokered access, scoped policy, audit logs, and targeted revocation. - [What Is a Secrets Broker for AI Agents? — Gazebo](https://gazebohq.com/writing/secrets-broker.md): A secrets broker checks access policy, returns a scoped token instead of the raw key, and logs every request. How it differs from a secrets manager. - [Why AI Agents Shouldn't Share API Keys — Gazebo](https://gazebohq.com/writing/agent-iam-101.md): Sharing API keys with AI agents gives them root access to your account. Here's the blast radius problem and how to fix it with IAM. - [MCP Security: What Developers Need to Know — Gazebo](https://gazebohq.com/writing/mcp-security.md): MCP security for developers: shared token risks, prompt injection via MCP tools, per-agent bearer tokens, and what a secure MCP setup looks like. - [How to Set Up a Cursor Agent with Scoped Service Access — Gazebo](https://gazebohq.com/writing/cursor-agent-setup.md): Set up your Cursor agent with scoped service access via Gazebo MCP. Step-by-step guide — no raw API keys in your config. - [AWS Secrets Manager for AI Agents: IAM Roles vs Tokens — Gazebo](https://gazebohq.com/writing/aws-secrets-manager-ai-agent-workflows.md): AWS Secrets Manager with IAM roles works for AWS-native agents. For agents crossing service boundaries, scoped tokens fill the IAM gap. - [Pasting API Keys into AI Agent Prompts: The Risk — Gazebo](https://gazebohq.com/writing/pasting-api-key-into-agent-prompt.md): Pasting an API key into an agent prompt puts it in conversation logs and provider infrastructure. Here's where the key goes and how to stop it. - [API Key Rotation vs. Revocation for AI Agents — Gazebo](https://gazebohq.com/writing/api-key-rotation-vs-revocation.md): API key rotation breaks every system sharing a key. Revocation cuts one agent's access without touching the credential. Why agents need revocation. - [OAuth 2.0 for AI Agents: Client Credentials Explained — Gazebo](https://gazebohq.com/writing/oauth-2-for-agents.md): OAuth 2.0 client credentials give broad static access — fine for services, wrong for agents. Agents need per-agent scoped tokens and revocation. - [HashiCorp Vault and Gazebo for AI Agent Access — Gazebo](https://gazebohq.com/writing/hashicorp-vault-vs-gazebo.md): Vault handles infrastructure secret storage. Gazebo adds per-agent approval gates and audit trails on top. How to layer them without changing your Vault setup. - [Least Privilege for AI Agents: A Practical Guide — Gazebo](https://gazebohq.com/writing/least-privilege-ai.md): Least privilege for AI agents means scoped credentials, audit logs, and revocation per agent. Here's how to implement it without rebuilding your stack. - [HashiCorp Vault Agent and AI Credentials — Gazebo](https://gazebohq.com/writing/vault-agent.md): Vault Agent handles auto-auth and secret injection for infrastructure. How it fits with the AI agent credential layer and how both stack together.