THREAT DETECTION

ML-powered protection against AI threats

Real-time detection of prompt injection, jailbreak and attack-pattern retrieval, PII leaks, data exfiltration, toxicity, fraud & abuse, secret keys, API keys, malware, URL threats, tool injection, malicious entities, code vulnerabilities, gibberish, and language policy. 15 detectors (10 always-on, 5 opt-in), backed by ~1,000+ patterns out of the box.

Start for Free View Documentation

THE PATH A PROMPT TAKESfig. 01

6 layers, in sequence, ahead of the model. Any one of them can end the request — they all write to one decision, with the reason that produced it.

What each layer checks

Prompt Injection Detection: ML-powered classification detects injection attempts including instruction override, role manipulation, context breaking, and jailbreak attempts.
PII Detection & Protection: Detect and protect 43 entity types of personally identifiable information across 11+ countries with checksum validation.
Data Exfiltration Blocking: Detect and block attempts to extract system prompts, training data, or other sensitive information.
Toxicity Filtering: Block harmful, inappropriate, or policy-violating content with configurable severity thresholds.
Secret Key Detection: Automatically detect and redact API keys, secrets, and credentials with entropy analysis before they reach the LLM.
URL Filtering: Detect and block malicious, phishing, or unauthorized URLs in prompts and responses to prevent data exfiltration via external links.
Fraud & Abuse Prevention: Identify and rate-limit automated abuse, bot attacks, and fraudulent behavior through behavioral analysis and request fingerprinting.
Malware Detection: Block malicious code, destructive commands, and potentially harmful instructions before they reach the model.
Jailbreak Detection (LLM): LLM-powered jailbreak detection catches sophisticated bypass attempts that evade traditional pattern matching, including multi-turn and encoded attacks.
Tool Injection Detection: Detect and block attempts to inject malicious tool calls or manipulate agent tool usage through crafted prompts.

How Threat Detection Works

01
Intercept
Every request passes through PromptGuard's security layer before reaching your LLM provider.
02
Analyze
ML models and ~1,000+ detection patterns analyze the request across 15 detectors (10 always-on, 5 opt-in). Deterministic checks clear most traffic in under 10 ms; requests escalated to ML or an LLM judge are network-bound and slower.
03
Protect
Malicious requests are blocked, logged, and alerted. Safe requests pass through unmodified.

Zero-Config Protection

python

from openai import OpenAI

# Just change your base URL - that's it!
client = OpenAI(
    base_url="https://api.promptguard.co/api/v1",
    api_key="your-openai-key",
    default_headers={
        "X-API-Key": "pg_live_xxxxxxxx"
    }
)

# All requests are now protected
response = client.chat.completions.create(
    model="gpt-5-nano",
    messages=[{"role": "user", "content": user_input}]
)

# Malicious prompts are automatically blocked
# No code changes needed!

Why PromptGuard Threat Detection?

15 detectors (10 always-on, 5 opt-in) out of the box
ML-powered, not just regex matching
Sub-10 ms fast path, opt-in slower tiers
Streaming support for all providers
10,000 free requests/month

Other controls on the same path

Red Team Testing

Test your defenses with adversarial attacks.

AI Agent Security

Secure autonomous AI agents.

PII & DLP

Protect sensitive data in prompts.

Protect Your AI Application

Start blocking prompt injection and other threats in under 2 minutes. No code changes required.

Get Started Free Contact Sales