# Latent Injection via WHOIS Domain Record Data

- **Severity:** High
- **Signature ID:** `3f91dba2-ca88-4363-baf8-e3ce73381483`
- **MITRE ATLAS:** AML.T0051 (LLM Prompt Injection), AML.T0051.000 (Direct)
- **OWASP:** ASI01 (Agent Authorization and Control Hijacking), LLM01 (Prompt Injection)

## Summary

Attackers hide malicious instructions inside WHOIS records — the technical details returned when looking up who owns a domain. When an AI agent summarizes or reads that WHOIS data, it can be tricked into following the hidden instructions instead of doing its actual job.

## How the attack works

An attacker registers or crafts a domain whose WHOIS record fields (like Registry Domain ID, Registrar WHOIS Server, Registrar URL, or creation/update dates) contain text formatted to look like normal WHOIS output but actually carries prompt injection commands. Someone asks an AI agent to look up or summarize that domain's WHOIS information. The agent fetches the WHOIS record and treats the entire text, including the hidden instructions, as trustworthy input. The injected text can then override the agent's original task, making it leak data, take unintended actions, or ignore its instructions.

## Why it matters

An organization using an AI agent to process domain intelligence or WHOIS lookups can have that agent hijacked simply by pointing it at a malicious domain, without any direct access to the agent's systems.

## What you can do

- Treat WHOIS output as untrusted data, not as instructions, when feeding it to an AI agent.
- Strip or sanitize WHOIS fields before passing them to a language model, especially free-text fields like registrar names, dates, and IDs.
- Use dedicated WHOIS parsers to extract only the specific fields needed, rather than passing raw WHOIS text to the agent.
- Review agent logs for cases where domain summaries include unexpected instructions, links, or behavior changes.

## Known benign look-alikes

- Legitimate WHOIS output pasted by a user for domain research (low risk, no embedded injection payload)
- Security tools that process WHOIS data in isolated sandboxed contexts
- Domain registrar API responses processed by purpose-built WHOIS parsing tools

## References

- https://agentthreatrule.org/en/rules/ATR-2026-00493

---
Source: https://www.netzilo.com/threats/atr-latent-injection-via-whois-domain-record-data