AI agents
Software that acts as you.
An AI assistant connected to your business is a new user with your credentials. It sends mail as your domain. It may hold keys you would not give a new employee on their first day. And one that reads your inbox is following instructions written by anyone who can email you.
None of that is a reason to avoid them. It is a reason to know what your domain currently allows before you connect one.
Free. No sign-up. About ten seconds.
Where an agent touches your domain
Four of them, and each ends at a record you can check today.
It becomes a sender you never authorised
An assistant wired to a mail service sends as your domain. If that service is not in your SPF record, those messages fail authentication.
The usual "fix" is to loosen SPF until everything passes, which is switching the lock off because you kept losing the key.
SPF
It pushes SPF past the limit nobody sees
SPF allows ten DNS lookups. Your mail platform, newsletter, CRM and invoicing each cost about one. An agent can be the eleventh.
Past ten, receivers stop evaluating and SPF fails for every message you send — including the ones that worked yesterday. Nothing bounces loudly.
SPF lookup count
It reads mail, so it takes instructions from strangers
An agent reading your inbox cannot reliably separate text you wrote for it from text somebody emailed it. Your address is published; anyone can write to it.
Part of the defence is not in the AI at all: whether a message can be shown to be genuine is decided by SPF, DKIM and DMARC.
DMARC
It holds keys somebody issued in a hurry
An agent with a broadly scoped DNS API token can rewrite your zone. That is not an application risk; it is the whole domain.
Scope the token to what the job needs, keep it out of the repository, and know who can revoke it at 2am.
Not a record — an access decision
The order that works
Before you connect anything to email: check what your domain publishes, get DMARC to enforcement carefully — starting at p=none with reports, and reading them for a few weeks before tightening — and confirm your mail platform acts on other domains' policies, which is what protects your inbox rather than everyone else's.
When you connect it: add its sending service to SPF deliberately and re-count the lookups, decide what it may do without a human, give it its own scoped credentials, and read its logs occasionally. An agent doing something strange on the fourth Tuesday is only visible if somebody looks.
What none of this fixes
Authentication stops someone sending as a domain they do not control. It does not stop someone registering a lookalike domain and sending from that — correctly authenticated, and owned by an attacker. Nor does it stop a real account being compromised and used to send real mail with bad instructions in it.
So this removes one whole class of attack, the free and no-skill-required class, and leaves the harder ones. That is what any honest security measure does.
Reading
- An AI agent in your inbox obeys whoever emails you
Connect an assistant to your email and every message becomes an instruction it might follow. Part of the defence is not in the AI — it is in your DNS.
- AI in Kenya Is About Workflow, Not Hype
Where AI is already creating real value for SMEs, and where the conversation has gotten ahead of itself.
The layer underneath, in full: domain security.