Skip to content
USE CASE: AI AGENTS

Secure autonomous AI agents

Your AI agents execute tools, browse the web, and make decisions. PromptGuard ensures every action is safe, validated, and auditable.

Start for FreeView documentation
THE ATTACK A PER-MESSAGE FILTER MISSESschematic
user objectivestated once, at turn 1agent tracesearchopen docopen doclist keyshttp postaudited against the objective, across the whole traceevery one of them clears the content filter on its ownheld for reviewfail-closedan unrenderable verdict holds
Each action is clean on its own; together they are not the job you asked for. The audit reads the latest action against your original objective across the whole trace, and it is fail-closed, so a verdict it cannot render holds the action rather than passing it.

Where an autonomous agent is exposed

  1. 01

    Tool Execution Validation

    Every tool call is validated before execution. Block dangerous operations, detect injection attempts, and enforce least-privilege access.

  2. 02

    Web Browsing Protection

    When your agents scrape the web, we scan content for hidden instructions and indirect prompt injections.

  3. 03

    Behavioral Baselines

    We learn your agent's normal behavior patterns and alert you to anomalies that could indicate compromise.

  4. 04

    Human-in-the-Loop

    Require human approval for sensitive operations like financial transactions, data modifications, or external communications.

  5. 05

    Session Isolation

    Each agent session is isolated. Compromised sessions can't affect other users or escalate privileges.

  6. 06

    Complete Audit Trail

    Every decision, tool call, and action is logged. Full visibility for compliance and incident response.

Why PromptGuard for AI agents?

  • Purpose-built for autonomous agents
  • Deep integration with agent frameworks
  • Behavioral anomaly detection
  • Human-in-the-loop workflows
  • Included in all plans

How it works for AI agents

  1. 01

    Deploy

    Integrate PromptGuard into your agent framework. Works with LangChain, AutoGPT, CrewAI, and custom agents.

  2. 02

    Protect

    Every LLM call and tool execution is secured. Threats are blocked, sensitive operations require approval.

  3. 03

    Monitor

    Real-time dashboards show agent activity, security events, and behavioral anomalies.

Securing LangChain agents

python
import subprocess
from promptguard import PromptGuard

pg = PromptGuard()  # reads PROMPTGUARD_API_KEY

def guarded(tool_name, arguments, run):
    # PromptGuard validates the call before it executes
    check = pg.agent.validate_tool(
        agent_id="support-agent", tool_name=tool_name, arguments=arguments
    )
    if not check["allowed"]:
        # Dangerous calls are refused; others need human review
        raise PermissionError(check["reason"])
    return run(**arguments)

def execute_shell(command: str):
    return guarded("execute_shell", {"command": command},
                   lambda command: subprocess.run(command, shell=True))

The controls this sector runs on

Secure your AI agents today

Deploy autonomous AI agents with confidence. Enterprise-grade security from day one.

Get started freeContact sales