Greymatter.ioGreymatter.io
Start for Free
Greymatter.ioGreymatter.io Why Greymatter.io?
Pricing Start for Free
Product • AI Guardrails

Every AI tool, controlled and secured.

Already built in. No new tool required.

Production Platform 88 svc · 12 ns Filter Guardrails by namespace, service, rule or tag ▲ 10 degraded● 743 healthy
Guardrails Log ReportRecord ID: a7f3c0d2-●●●●-●●●●-91e1
SOURCE
greymatter.guardrails
NAMESPACE
assistant
SERVICE
chat-api
EVENTS
Oct 21, 2026
highest_severity 4mode enforceaction blockis_interrupted true
INTERPRETATION
APPGUARD'S INTERPRETATION OF WHAT HAPPENED
A hidden instruction told the model to ignore its rules. Guardrails stopped it before the model answered.
RULESETS · 1 ACTIVE IN THIS RECORD
● promptinjectiondisclosurephisupplychainpoisoningoutputhandlingagencypromptleakageembeddingmisinformationconsumption
Verified, not claimed
Every OWASP AI risk, covered
All ten OWASP LLM Top 10 categories.
See the 10/10 →
One platform, not a new tool
Uses the identity and policy in place.
See Identity →
A person behind every finding
Each alert shows who sent the request, when they are signed in.
See Forensic Auditing →
Running AI Guardrails

Real safeguards, not guesswork, around every AI agent.

Who acted. What is allowed. What gets stopped.

01Attribution 02Policy 03Enforcement

No agent action is anonymous.

✓When an AI agent oversteps, the record points to a real person, not a log entry
✓Even when an agent is acting on someone's behalf, that person stays responsible, not just the agent
Production Platform 88 svc · 12 ns Search identities, request ids, namespaces ▲ 10 degraded● 743 healthy
230K
AUDIT EVENTS · 24H
0
WAF BLOCKED
0
POLICY DENIED
5
IDENTITIES SEEN
AUDIT EVENTS BY NAMESPACE · SCOPE
orders166K
payments35K
records17K
billing4,591
reports1,306
HOW IDENTITY ARRIVEDrequests at the first audited hop
Service to service a SPIFFE source185,958 · 81%
Passed on by an earlier hop x-user-id35,873 · 16%
No identity refused at the hop7,723 · 3%
OWASP LLM Top 10
10/10

All ten OWASP AI risks, covered by one product.

Security teams often evaluate a separate tool for each new AI risk. Greymatter.io already covers it in one place. Nothing new to install.

LLM01 LLM02 LLM03 LLM04 LLM05 LLM06 LLM07 LLM08 LLM09 LLM10
threat file · LLM01 covered
The threat
Prompt Injection

A hidden instruction trying to hijack an agent gets caught before the model ever responds to it.

The Greymatter.io response Enforceable

Greymatter.io can block the request in-line, before the model or the user ever sees the threat. Enforcement turns on per module, after a team confirms accuracy in detect mode.

Guardrails · Engine On In the path · blocks
Analyst · /v1/chat/completions live traffic
Guardrails inspects ↓
RULE 9001 · CRITICAL · LLM01 Prompt Injection
BLOCKED
stopped in-line · forensic event written
Select a category to open its threat file. ←Swipe to change rule→
See Greymatter.io inspect outbound and inbound traffic.

Your own agents, your own traffic, checked from the first request.

See Greymatter.io inspect outbound and inbound traffic.

Your own agents, your own traffic, checked from the first request.

Related resources

Where this connects across the platform

Product · Identity
Identity
Learn More →
Product · Forensic Auditing
Forensic Auditing
Learn More →
Product · Web Application Firewall
Web Application Firewall
Learn More →
Swipe for more
FAQs
Questions worth answering directly

Greymatter.io AI Guardrails checks AI prompts, responses, MCP calls, and images against all 10 OWASP LLM Top 10 risks, plus tool misuse. Data loss protection flags personal data and classification markings, and never blocks.

No. Greymatter.io starts every new rule in detect mode, logging what it would have stopped without interrupting the request. Greymatter.io only blocks once a team explicitly turns enforcement on.

Yes. Greymatter.io AI Guardrails flags excessive agency and tool misuse in agent and MCP traffic. Run it in detect mode first, then turn on blocking.

Yes. When an AI agent reaches an MCP tool through Greymatter.io, the agent finds how to sign in and registers with your identity provider. The sign-in asks only for the scopes that tool needs, with no shared keys or copied credentials.

Yes. Greymatter.io AI Guardrails reads MCP tool calls, including live event streams.