Security is important to us.

AI agents are high-value targets. Their inboxes need to be safe not just from spam — but from adversarial content designed to manipulate them.

Prompt Injection Firewall

Emails are a natural vector for prompt injection attacks against AI agents. agentmbox scans inbound content for patterns designed to hijack agent reasoning — hidden instructions, Unicode tricks, and instruction-smuggling payloads — before they reach your agent's context window.

Hidden instruction detection

White-on-white text, zero-width characters, and CSS-hidden content are stripped and flagged before delivery.

Instruction-smuggling patterns

Known prompt injection templates and LLM jailbreak patterns are matched and neutralised in the message body.

Sanitised plain-text extraction

When agents request text extraction, HTML is stripped to safe plain text — removing scripts, style tags, and obfuscation layers.

Spam Filter

Every inbound message passes through multi-layer spam analysis before it reaches your agent's inbox. Scoring combines reputation checks, content heuristics, and sender authentication (SPF, DKIM, DMARC). Messages that fail are quarantined — not silently dropped — so you can audit what was blocked.

Sender reputation

IP and domain reputation is checked against real-time block lists before a connection is accepted at the SMTP layer.

Authentication enforcement

SPF, DKIM, and DMARC are validated on every inbound message. Failures are logged and scored against delivery.

Content scoring

Message body and headers are analysed for known spam patterns. High-scoring messages are quarantined automatically.

Abuse Filter

Outbound mail is rate-limited and monitored per mailbox to prevent abuse of the sending infrastructure. Agents that trigger volume or pattern anomalies are throttled and flagged for review. This keeps deliverability high for everyone.

Per-mailbox rate limiting

Sending volume is capped per wallet address. Burst attempts are queued and smoothed — not silently dropped.

Anomaly detection

Unusual sending patterns — sudden volume spikes, high bounce rates, repeated identical payloads — trigger automatic throttling.

Bounce & complaint handling

Bounce and spam-complaint signals are fed back in real time. Persistent abuse results in sending suspension.

Ready to give your agents email?

Start with $10 in free credits — no wallet or crypto setup needed.