Prompt injection monitoring for AI agents

Metorial Protoguard monitors every tool call, message, and execution for prompt injection attacks in real time. Attacks on your agents are detected before they do damage.

Get startedBook a demo
platform.metorial.com
A screenshot showcasing the Prompt Injection Monitoring feature of Metorial.

Prompt Injection Monitoring

Catch attacks aimed at your agents

AI agents introduced a new kind of attack: prompt injection, where hidden instructions try to trick an agent into doing something it shouldn't. Metorial monitors for these attacks across every interaction, using Protoguard, its central prompt security system, to inspect what flows through your agents in real time. A malicious instruction is caught rather than quietly obeyed.

  1. 01

    Monitor for prompt injection.

    Metorial watches integrations and skills for prompt injection attacks. This threat doesn't go unnoticed.

  2. 02

    Detect in real time.

    Protoguard inspects every tool call, agent message, and execution against a set of rules in real time, alongside your agents rather than in their way, to detect attacks as they happen.

  3. 03

    Use integrations with confidence.

    Because every interaction is monitored, teams can rely on the MCP servers they use, including third-party ones, knowing attacks are being watched for.

Part of your defenses

Protection across the whole path

Prompt injection monitoring works alongside Metorial's other protections to defend agent traffic end to end.

Gateway

Inspect everything

Monitoring runs on traffic through the Metorial Gateway. All MCP connections are covered.
Network

Contain the damage

Pair detection with firewalls and enclaves so a flagged interaction can be contained.
Alerts

Respond fast

Configure monitors and alerts so the right people are notified when an attack is detected.

How it protects

Monitoring on every interaction

  1. Tool calls and messages monitoredEvery
  2. DetectionReal-time and out of band
  3. Powered byMetorial Protoguard
  4. CoverageFirst-party, custom, remote, Docker

Watch every tool call. Detect injection in real time. Flag the attack. Keep agents safe.

FAQ

Answers to common questions about prompt injection monitoring and how they fit into governed AI agent infrastructure.

  1. What is prompt injection monitoring?
    It's monitoring for prompt injection attacks, where hidden instructions try to make an agent act against your intent. Metorial inspects agent interactions to detect these attacks before they cause harm.
  2. Protoguard is Metorial's central prompt security system. It monitors tool calls, messages, and executions in real time to detect prompt injection and other exploits.
  3. Every tool call, agent message, and integration execution is monitored against a set of rules designed to detect different kinds of attacks on agents.
  4. First-party integrations, custom MCP servers, remote MCP servers, and Docker MCP servers.
  5. Yes. You can configure monitors and alerts for prompt injection attacks. The right people know immediately.
  6. It works alongside the Metorial Gateway, firewalls, and enclaves. Detection and containment work together to protect agent traffic.

Defend your agents against prompt injection

See how Metorial Protoguard monitors every interaction to catch attacks aimed at your agents. Book a demo, or talk to sales.

Book a demoTalk to sales