Guides
Guides
This series leads with data control and sovereignty: you decide what an agent may reach. Progressive access — an address for the agent, a human who Allows — and an honest limit on what that protects. Prompt injection appears only as one reason control helps, never as the main brand story.
These are evergreen guides — category teaching, not news. When we claim a protection, we mean what Mail actually does in v1: a per-agent mailbox, a trust wall (Unknown → Pending → Allowed / Blocked), Allowed-only MCP/API reads, and no agent escalation. Not a jailbreak shield. Not a CASB.
Give your AI an address, not access to your inbox.
Pilots
Give your AI an address, not your inbox
Category manifesto. Why treating the inbox as an API is a bad default, how progressive grant works, and what Mail v1 is — and isn’t.
You control what data reaches your agent
Data control first. Prompt injection is one reason a trust wall helps (blast-radius reduction) — not the homepage promise. Honest: not a full LLM jailbreak / model-alignment shield.
Why “just connect Gmail to your agent” is a security smell
Smell ≠ never use OAuth. Smell = starting with maximum privilege. When OAuth still wins, and the progressive alternative.
Coming soon
- Progressive access vs all-or-nothing OAuth
- What to Allow first when onboarding an agent
- Audit: “what did my agent see?”
- Agent mailboxes vs AI email assistants
- Keep the agent. Keep the boundary. (operator’s field guide)
Tell your agent: start using Mail