PII Detection
An SSN in a prompt or an agent's reply. Caught and redacted.
SUPERWISE® AMP → Guardrails
The AI guardrails platform that blocks failures before they reach users
Enterprise AI guardrails are runtime product controls that evaluate live AI requests and enforce policy before a response reaches a user.
What the AI guardrails platform does
Framework guardrails define expectations, and code guardrails protect one application path. SUPERWISE runtime checks apply policy consistently across governed AI traffic.
Guardrails are the runtime layer of the SUPERWISE AI governance platform. Each request and response routed through the configured guardrail is checked against its configured rules for PII, toxicity, restricted topics, and jailbreak attempts. A detected violation is blocked before the user sees it.
On the platform, guardrails check PII, toxicity, topics, and jailbreak attempts on inputs and outputs. Through Sentinel, the AI gateway, the AI tools your team already uses get the secrets, PII, and toxicity guardrails on outgoing prompts.
See it work
Guardrails in action
superwise-guardrail-monitor
● Monitoring active...
Real-time policy checks
Built-in protections
Pre-configured guardrails for common risks, plus the flexibility to define your own
Toxicity Filter
Your agent just insulted a customer. Prevent harmful responses in real time.
Topic Restriction
User tried to make your agent discuss competitors. Topic blocked.
Jailbreak Detection
Attacker tried to override system prompt. Attempt logged and rejected.
Competitor Check
Agent almost named a competitor. Blocked before the word left.
Protection at every layer
1
Input Validation
Detect prompt injection, PII, and malicious inputs before model execution.
2
Runtime Evaluation
Evaluate AI outputs against your policies in real-time. Block, modify, or flag violations automatically.
from superwise_api.models.guardrails.guardrails import ToxicityGuard
high_toxicity_guard = ToxicityGuard(name="High toxicity guard", tags=["input"], threshold=0.8, validation_method="sentence")
results = sw.guardrails.run_guardrules(tag="input", guardrules=[high_toxicity_guard], query="Heck ye")
for result in results:
if not result.valid:
print(f"Guardrail '{result.name}' was violated.")
print(f"Violated message: {result.message}")Guardrails stop threats in real-time
Watch SUPERWISE detect and block PII leaks before they ever reach the LLM. See how a patient SSN gets stopped cold.
AI guardrails questions
What buyers ask when comparing AI guardrails
AI guardrails are runtime checks that enforce policy on AI inputs and outputs. Each request is checked before the response is returned, so a violation is blocked as it happens instead of being found in a later review. The block is logged, so you have the evidence too.
The built-in guardrails cover PII detection, toxicity, topic restriction, jailbreak detection, and competitor checks. Alongside them, you can add custom policy rules for your own restricted topics, brands, and content, and apply them across every model and agent.
Guardrails run inline: each request is checked before the response is returned, on inputs, outputs, or both, depending on how the guardrail is configured. The SUPERWISE console shows the average latency of each Sentinel, so you can see what your configured checks cost in practice.
A guardrail is a check, such as PII or jailbreak detection, applied to a prompt or a response. A gateway is where checks run on the traffic that passes through it. Sentinel, the SUPERWISE AI gateway, applies the secrets, PII, and toxicity guardrails to outgoing prompts from the AI tools your team already uses. On the platform, guardrails check the inputs and outputs of the agents you build. Both are the runtime layer of SUPERWISE AMP, and the same policy that blocks a request produces the audit trail.
Questions buyers ask
All questions- Are your AI controls enforced, or just written down?Usually written down. A policy document says what people are supposed to do; an operating control produces evidence that it prevents, detects, escalates, constrains, or corrects risky behavior.
- What does AI enforcement at runtime actually look like?One point that every AI request passes through on its way to a model, where rules are applied before a response reaches anyone.
- What is the OWASP Top 10 for LLM applications?No single security tool covers the OWASP Top 10 for LLM Applications. Coverage requires layered controls across prompts, model access, data and retrieval, outputs, permissions, dependencies, monitoring, and incident response…
Talk to us about your first guardrail
Turn your policy into runtime checks for the AI traffic you govern.