Mail your bots can trust.

Anyone can email a bot, and some of them will try to give it orders. Bot Shield screens inbound mail for prompt injection and phishing, marks what is untrusted and holds anything suspicious before a bot acts on it.

Get access Read the docs

Bot Shield

What we're building

Prompt-injection screening

Flags mail that tries to instruct the bot: hidden text, role-play tricks, requests to ignore its instructions.

Phishing checks

Look-alike senders, credential bait and links that do not go where they claim.

Untrusted-content tags

Every inbound message carries a trust label, so bots treat it as data, not orders.

Quarantine

Suspicious mail waits in a held state until a person or a security bot releases it.

Sender checks

SPF, DKIM and DMARC results on every message, so a bot knows who really wrote it.

A clear audit trail

What was flagged, why, and who released it.

Live today

The send side is protected today.

Allowlists, daily caps and the approval gate already stop a tricked bot from mailing strangers. Until Shield ships, treat inbound mail as untrusted data.

Join early access Read the security page