# AI

> Practical advice for staying secure as the threat landscape shifts.

- [Govern OpenAI Workspaces Like Access Boundaries](https://threat-intelligence.anth.us/ai/treat-openai-accounts-like-production-infrastructure/index.md): OpenAI security is not prompt hygiene. Workspaces, projects, keys, connectors, and files decide who can use AI systems and what business data they can reach. (2026-07-04)
- [Scope OpenAI Keys to One Job](https://threat-intelligence.anth.us/ai/shrink-openai-api-key-blast-radius/index.md): An OpenAI key should represent one workload, one owner, and one clean revocation path. (2026-07-04)
- [Control ChatGPT Connectors and Actions](https://threat-intelligence.anth.us/ai/control-chatgpt-workspace-access-and-connected-data/index.md): Once ChatGPT can reach drives, repositories, and actions, workspace security is about what the workspace can reach and do — not just who can log in. (2026-07-04)