Security
12 posts on security for AI agents that send and receive email.
6 August 2026 · 4 min read
Rotating an agent's key without breaking it
Overlap windows, where keys actually leak from, what to do when one is exposed, and why the agent should never be the thing that holds the only copy.
5 August 2026 · 4 min read
Working out who actually sent that message
The From header is a claim. What authentication results actually prove, how to check a sender against what you already know, and why identity should never authorise an action.
4 August 2026 · 4 min read
Sending the model less than you think it needs
Every message you pass to a model is data leaving your systems. What to strip, what to summarise, what never to send at all, and why the cheapest version is usually the safest.
3 August 2026 · 4 min read
Every tool you connect is a tool your inbox can be talked into using
An agent with an inbox and a dozen connected tools has a supply chain. What a hostile or compromised server can do, why email raises the stakes, and how to connect one safely.
2 August 2026 · 5 min read
When someone sends email pretending to be you
DMARC enforcement stops spoofing of your exact domain and does nothing about lookalikes. What each defence actually covers, what to monitor, and what to do the day it happens.
2 August 2026 · 5 min read
Deciding what each agent is allowed to see
Least privilege inside a workspace: what an identity can reach by default, how contact, note, and vault grants work, which human roles can change them, and how to design the smallest useful agent.
31 July 2026 · 5 min read
When the agent sends something it should not have
A runbook for the hour after an agent sends the wrong thing: stopping it, working out the blast radius, telling the recipients, and the postmortem that stops it repeating.
31 July 2026 · 5 min read
Where the mail actually goes: data flow for an agent inbox
The question procurement asks and most teams cannot answer: what leaves your infrastructure when an agent reads and sends mail, who processes it, and which parts you control.
29 July 2026 · 5 min read
The security review questions an email agent should survive
The questions a security team will ask before an agent touches customer mail, why each one is asked, and what a good answer looks like when the agent is the thing under review.
29 July 2026 · 5 min read
Keys, scope, and what to do the day one leaks
An agent key is a credential that can send email as your company. How to scope, store, rotate, and revoke keys, and the hour-one checklist for a key that has leaked.
27 July 2026 · 5 min read
Login codes, 2FA, and verification email an agent has to handle
Agents that sign up for services hit email verification and two-factor prompts. How to receive codes in an agent inbox, clear TOTP without a human relaying digits, and not build a phishing target.
26 July 2026 · 8 min read
Prompt injection by email, and how to build an agent that survives it
An email inbox is an untrusted input channel anyone can write to. The injection patterns agents actually get hit with, and the architectural defences that work better than telling the model to be careful.