Skip to content
Guardino Sentinel

The AI era opened a new attack surface.

Sentinel catches prompt injection, leaked API keys, and the threats born with generative AI in real time, right in your browser. Nothing you paste ever leaves your machine.

Built for the engineers and AppSec teams putting LLMs into production.

Try the live firewallExplore the threat map

Private by design · OWASP-aligned · No account needed

6-Layer Defense Pipeline & Attack Lab

Watch all 6 Sentinel engines intercept attacks in under 10 milliseconds.

Select a real-world enterprise leak or adversarial scenario below. Every stage runs deterministically in your browser and produces a KMS-signed SOC2 / KVKK audit receipt with zero raw data retention.

Real-World Attack Scenarios

Live 6-Core Engine Telemetry Pipeline

Total Pipeline Latency3.62ms (<10ms SLA)
Final VerdictWARN (90/100)
  1. 1. OWASP LLM Top-10 Prompt Injection
    0.32 msPASS
    Instruction override, jailbreak & delimiter escape analyzerOWASP LLM01 clean
  2. 2. High-Entropy Secret Scanner
    0.58 msREDACTED
    Shannon entropy & structural prefix redactor ([REDACTED_AWS_KEY])1 secret(s) -> [REDACTED_SECRET_KEY]
  3. 3. Checksum PII Redactor
    0.41 msREDACTED
    Mod-10/11 TCKN, ISO-7064 IBAN & Luhn payment card maskerEMAIL masked
  4. 4. Zero-Width Canary Trap
    0.15 msPASS
    Invisible Unicode watermark & exfiltration tripwireCanary guard standby
  5. 5. Output Leak Verifier
    0.22 msPASS
    Real-time completion stream & system prompt leak blockerCompletion stream clean
  6. 6. GLiNER + NeMo Semantic DLP & KMS Audit
    1.94 msSIGNED
    Zero-shot trade secret NER (<10ms) & SHA-256 hash-chained ledgerKMS SHA-256 ledger signed
Sanitized Payload Sent to Model
Deploy production worker using SECRET_KEY_ID=[REDACTED_API_KEY] and notify [REDACTED_EMAIL] when active.
KMS-Signed SOC2 / KVKK Audit Receipt (Zero Raw Data)
prev_hash:
000000000000000000000000...
entry_sha256:
13110efa93a9ba3a8d2677431642...
kms_sig:
kms:rsassa_pss_sha256:241975acc049ce08ba2e0c1545a20463d54af9bc
SOC2 CC6.1ISO 27001 A.8.12KVKK Madde 12GDPR Art. 32

Deploy across your stack in 30 seconds

STDIO / MCP

MCP Server (Claude / Cursor / Windsurf)

Zero-config local stdio shield for AI coding agents.

npx -y @guardino/sentinel-mcp
MANIFEST V3

Chrome Extension (Ghost Mode)

Inline DOM protection for ChatGPT, Claude, Gemini & Copilot.

chrome://extensions -> Guardino Sentinel (Ghost Mode MV3)
REST / GATEWAY

Enterprise API Gateway

Drop-in proxy with <10ms GLiNER GPU DLP & KMS Audit Trail.

curl -X POST https://sentinel.guardino.ai/api/v1/shield -d '{"prompt":"..."}'
Live · Prompt Injection Firewall

Paste a prompt. Watch it get caught.

Direct and indirect injection, jailbreaks, system-prompt exfiltration, encoded payloads, classified against the OWASP LLM Top 10, instantly.

Or try a real attack:

Runs entirely in your browser. Nothing is sent anywhere.

Blocked
53
Risk score
  • HighInstruction overrideLLM01

    Ignore all previous instructions

Live · Secret & API-Key Scanner

Find the keys before attackers do.

Cloud keys, model API tokens, database URLs, private keys and high-entropy secrets, caught with provider fingerprints and entropy analysis.

Or load a sample:

Runs entirely in your browser. Nothing is sent anywhere.

4 secrets found
  • HighDatabase URL with credentialsLine 2

    post••••••••••••/app

    Move credentials to a secret manager; rotate the password.

    Verified pattern
  • CriticalStripe secret key (live)Line 3

    sk_l••••••••••••EFGH

    Roll the key in the Stripe dashboard immediately.

    Verified pattern
  • CriticalGitHub tokenLine 4

    ghp_••••••••••••6789

    Revoke the token in GitHub settings and rotate.

    Verified pattern
  • MediumJSON Web TokenLine 5

    eyJh••••••••••••0000

    Treat as a bearer credential; confirm it is not long-lived.

    Verified pattern
How it works

Three steps. Zero data leaves your browser.

  1. 01

    Paste

    Drop in a prompt, a tool result, source code or a config file: anything untrusted.

  2. 02

    Analyze locally

    Deterministic rules, provider fingerprints and entropy run in your browser. No network call, nothing stored.

  3. 03

    Get an OWASP verdict

    A clear verdict, a risk score and the exact technique: mapped to the OWASP LLM & API Top 10.

The Threat Observatory

Every new-age threat, on one map.

The AI-era attack surface, charted against OWASP and industry frameworks. Two are live today, the rest are lighting up.

Live demoActive detectionRoadmap
AI & LLM

OWASP LLM Top 10 · 2025

APIs

OWASP API Top 10 · 2023

Secrets & Supply Chain

GitGuardian · SLSA

Human & Org

Shadow AI · Compliance

Live nowLLM01

Prompt Injection

Untrusted input rewrites the model's instructions, directly or smuggled through data the model reads.

73%

of production AI deployments hit by prompt injection

Source: AI Security Statistics 2026, Practical DevSecOps
Why now

The numbers moved faster than the defenses.

73%

of production AI deployments hit by prompt injection

Source
28.6M

new hardcoded secrets pushed to public GitHub in 2025

Source
76%

of organizations now cite shadow AI as a real risk

Source
+81%

year-over-year surge in leaked AI-service credentials

Source

Figures are from third-party industry research, linked at the source. Guardino publishes no detection-rate or SLA claims it cannot prove.

Why Sentinel

Built for how the AI era actually breaks.

Private by design

Detection runs in your browser or at the edge. Your prompts and code never sit on our servers.

Real-time, not after the breach

Findings appear as you type, fast enough to live inside an editor, a CI step, or a chat box.

Mapped to OWASP

Every verdict is tied to the OWASP LLM and API Top 10, so your team speaks one language.

Featherweight

No agent, no heavyweight model to host. The core is a few hundred kilobytes of deterministic logic.

AI against AI

Pro adds an optional semantic layer powered by Claude for the attacks rules alone can't see.

Developers

Ship it into your stack.

The same engine that powers the live tools: as an SDK, a REST API, and a CI gate.

Call the detection engine from TypeScript.

import { analyzePrompt } from "@guardino/sentinel";

const result = analyzePrompt(userInput);
if (result.verdict === "blocked") {
  reject(result.findings); // OWASP-tagged techniques
}

SDK, API and CI integration ship with Pro; join the waitlist.

Pricing

Start free. Scale when you ship.

Free

$0

The live tools, forever.

  • Prompt injection firewall
  • Secret & API-key scanner
  • OWASP-aligned verdicts
  • Runs fully in-browser
Use it now
Most popular

Pro

$29/mo

For teams building with AI.

  • Everything in Free
  • Detection API & SDK
  • Claude semantic layer
  • CI/CD secret + injection gate
  • Scan history & alerts
Join the waitlist

Enterprise

Custom

For regulated scale.

  • Self-host & air-gap
  • SSO & RBAC
  • Custom rules & policies
  • EU AI Act / KVKK / GDPR support
  • Priority response
Talk to us
FAQ

Questions, answered.

Does my data leave the browser?

No. The firewall and scanner run entirely client-side — there is no network request and nothing is stored. You can confirm it in your browser's network tab.

What does it detect?

Prompt injection — direct, indirect, jailbreaks, system-prompt exfiltration, encoded and obfuscated payloads, and tool/agency abuse — plus leaked secrets across 15+ providers and high-entropy generic secrets, all mapped to the OWASP LLM & API Top 10.

Is it accurate?

The engine is deterministic and OWASP-aligned, hardened against spacing, leetspeak and base64 evasion. It favors precision: a lone destructive command is flagged, not blocked, to avoid false positives on legitimate developer questions.

Can I use it in my product?

Yes — Pro ships the same engine as an SDK, a REST API, and a CI gate for pull requests. Join the waitlist for early access.

Do the stats on this page come from Guardino?

No. Every figure is third-party industry research (OWASP, GitGuardian and others), linked at the source. We publish no self-measured detection-rate or SLA numbers.

What about the rest of the threat map?

Prompt injection, system-prompt leakage and secret leaks are live today. The remaining OWASP LLM/API and supply-chain categories are on the roadmap — the map shows the status of each.