AI APPLICATION SECURITY ASSESSMENT & RED-TEAM

Your AI stack. Every angle.
One auditable gate.

23 battle-tested security sensors orchestrated into one platform — from source code to LLM agents. Normalized findings, contextual risk, release gates, and evidence you can defend. Open-source sensors, proprietary core. Zero vendor lock-in.

Test It Live On This Site Start Scanning What You Get
aisg — scan pitchflow
LIVE
$ aisg scan my-app --profile pr
→ SAST (pattern rules)3 findings
→ SAST (data-flow)2 findings
→ SCA (dependencies)1 CVE
→ Infra & secrets1 critical
→ LLM (red-team)4 issues
✓ Release gate: BLOCKED (vibe-gate policy)
✓ Report: 24 sections + evidence + provenanceready

AISG IN NUMBERS

23Security Sensors
11Sensor Categories
20+Checklist Items
18Release Gate Steps
100%Audit Evidence

See it working. No signup.

Press one button and AISG assesses this very website in front of you — the public security surface: TLS, security headers, redirects, robots, sitemap and DNS. Results appear below in seconds, straight from our production endpoint.

This checks our site, not yours — we deliberately take no target input, so the demo can never be pointed at someone else's server.
Nothing running yet — press the button.

What a report actually says.

No score without evidence. Every AISG report is assembled from real findings, ordered by impact, with the limits written down. These numbers come from our own scans (archived 30 September 2026).

1. The summary you can act on

Risk score, severity breakdown and the exact sensor status — including the ones that did not run. Our repository scan: 4,418 files · 966,090 lines · 1,476 findings, split into 41 critical · 98 high · 405 medium · 932 low.

Sensor status is stated honestly: dependency scanners reported “skipped — no lock files found” rather than staying silent.

2. Findings with evidence, not adjectives

Each item carries the sensor, the rule, the severity, the file and line, and a validation state (unverified by default — we never present triage we did not do). CVE-backed findings are enriched with CISA KEV and EPSS probability.

Test fixtures and duplicate sensors are called out explicitly, so 1,476 findings do not read like 1,476 emergencies.

3. Remediation, coverage and governance

A prioritised fix plan with before/after code examples, a 20-item security checklist showing which controls were actually covered and which stay manual, plus an audit-grade export (Markdown + PDF).

AISG is layer-1 prevention, not a penetration test: nothing is exploited, and results require human review.

Specimen findings — from our own scans, sanitised
gitleakshighHardcoded secret (generic-api-key)config/** · redacted
promptmaphighPrompt stealing — system instructions extractedLLM pipeline · live probe
trivycriticalDependency CVE (bundled component)vendor bundle · line 1
zapmediumSource Code Disclosure - SQLhttps://aisg.web.id · verified false positive
The last row is shown on purpose: severity alone is not a verdict. We verify before we escalate.

Inside every report

Public-URL sample from the same session: 11 findings from httpx, dnsx, subfinder, nuclei and ZAP — mostly informational, including one verified false positive that we published instead of hiding.

How to read an AISG report How the pipeline works

Capabilities.

Not another scanner. A decision engine that orchestrates the best open-source sensors and turns raw findings into auditable security decisions.

01Orchestrated Detection

23 security sensors across source code, dependencies, secrets, infrastructure, web, and LLM agents — normalized into one schema. Sensors complement, never duplicate.

SAST · SCA · SECRETS · LLM · WEB

02Contextual Risk

Raw CVSS is not a decision. AISG scores with 8 risk factors — exposure, data classification, AI components — and tracks residual risk after controls.

RISK ENGINE

03Attack-Path Correlation

Findings across tools are correlated into attack chains with confidence levels — from indication to confirmed — instead of isolated alerts.

CORRELATION

04Policy-as-Code

Policy gates evaluate automatically: severity gates, secrets, missing controls, and the Vibe-Coding Gate block unsafe releases.

GOVERNANCE

05Release Gate & Provenance

18-step release gate, human approval matrix, generated-code provenance, and release manifest — evidence for auditors, not vibes.

AUDIT TRAIL

06Continuous Evaluation

Trend & delta tracking, version drift detection (model/prompt/RAG changes re-trigger evaluation), CI & pre-commit templates.

CLOSED-LOOP
SASTSCASECRETSINFRAWEBLLM RED-TEAMMCPRECONMISCONFIGAPI FUZZTLSEXPLOITABILITY+ 23 SENSORS · ONE GATE SASTSCASECRETSINFRAWEBLLM RED-TEAMMCPRECONMISCONFIGAPI FUZZTLSEXPLOITABILITY+ 23 SENSORS · ONE GATE

One Pipeline. Total Visibility.

From discovering assets to continuous monitoring — every stage leaves auditable evidence.

DISCOVERassets & AI-BOM
CLASSIFY8-factor risk
THREAT MODELSTRIDE·ATLAS·ATT&CK
CONTROL43 remediation playbooks
ASSESS23 sensors
AUTHORIZEhuman approval
GATErelease + vibe-gate
MONITORcontinuous eval

LLM Depth. Your Choice.

Scan your AI app — then go deeper. One optional field (your API key, encrypted, never leaves your machine) unlocks live red-team attacks on the model itself.

STANDARD — NO KEY

Full code-level assessment of your AI application. The core mission of AISG.

  • ✓ 23-sensor catalog — SAST, SCA, secrets, infra, DAST, API & asset, LLM (routed by target type)
  • ✓ Hardcoded API keys & secrets in your code
  • ✓ Insecure RAG flow & LLM integration (prompt-injection sinks)
  • ✓ Vulnerable AI libraries & outdated dependencies
  • ✓ Live app & API surface scan (DAST)
  • ✓ Layered reports: summary → governance → audit-grade
DEEP — WITH KEY (OPTIONAL)

Everything in Standard, plus LIVE red-team attacks against your model endpoint.

  • ✓ Everything in Standard
  • ✓ Jailbreak — model forced out of its rules
  • ✓ Prompt injection (direct + indirect / RAG poisoning)
  • ✓ Data leak / PII exposure & system-prompt extraction
  • ✓ Multi-turn attack chains + app-level input-path evals
  • ✓ MCP tool-abuse checks
  • ✓ Live evidence: attacks really executed against your model

API key is optional — leave it empty and everything still runs. Free/local models (Ollama, vLLM) need no key. Keys are encrypted at rest and never shown again. No key = LLM layer skipped transparently.

Transparent Risk Scoring.

AISG doesn't just throw a number at you. Every finding is weighted by severity, then the score is capped by the worst issue found — so a pile of low-severity notes can never masquerade as a critical breach.

Severity-weighted & deterministic

Every finding contributes to the score by severity — the more severe an issue, the heavier its weight. The result is a single score in the range 0–100 that is fully reproducible: re-scan the same code and you get the same score.

Anchored by the worst issue

A pile of low-severity notes can never masquerade as a critical breach. The score reflects both how many issues you have and how severe the worst one is — it only reaches the top of the scale when a truly critical issue is present.

Deterministic & auditable

Every sensor, rule and weight is pinned to a versioned engine — reproducible and verifiable on demand. Risk score is an indicator of finding count & severity, not a proof of exploitability.

Pricing.

From solo developer to enterprise — pick the depth you need. Your code & keys never leave your machine.

🎁
EARLY ACCESS — FREE for the first 25 developers

7 days FREE — Beginner-level scan: 2 targets · 4 scan runs (2 scans per target) · full report.
Fair use: one slot per account, max 2 claims per network.

Get Started

COMMUNITY

Rp 0for individuals
  • 2 targets · 2 scans/mo
  • SAST + secrets scanning
  • Findings count only
  • Local agent — code never leaves
  • Community support
Start Free

BEGINNER

Rp 50rb/month · solo developer
  • Everything in Community +
  • 5 targets · 50 scans/mo (local + URL share one quota)
  • 7 core sensors — multi-language SAST, deep-flow analysis, dependency & secrets scanning, IaC misconfig
  • LLM evals/mo — red-team your AI (model, multi-turn & app-level attack suite)
  • Risk score + severity breakdown report
  • Findings detail: tool, title, file:line
  • PDF report export — summary, risk score & detailed findings
Get Access

EXPERT

Rp 100rb/month · power user
  • Everything in Beginner +
  • 25 targets · 250 scans/mo (local + URL share one quota)
  • 23-sensor catalog — SAST, SCA, secrets, infra, DAST, API & asset, LLM (Local 10 · URL deep 7 · LLM 3, +MCP on demand)
  • Deep URL scan — active DAST crawl, API fuzzing, asset discovery
  • 3 parallel scans · faster results
  • Release gate · trend · PDF export — governance reports
  • Priority support
Get Access

ENTERPRISE

Customannual · SLA
  • Everything in Expert +
  • Unlimited scans & targets
  • Immutable audit trail + signed reports (SHA256)
  • Full audit report (.md) + PDF export
  • Runs entirely on your machine — code & keys never leave
  • Dedicated support
Contact

Pay securely via Virtual Account (BRI, CIMB Niaga, Permata, Mandiri) — choose your own bank and pay from its m-banking (inter-bank transfer also works). Beginner, Expert & Scan Pack are available now.

More payment options coming soon. Need more? Scan Pack — Rp 20rb / 15 scans · 2 targets (detailed findings + location, 30 days, after your tier quota).

FAQ.

Straight answers — including the one you're thinking about.

Why pay if the sensors are open-source?

Open-source = the sensor engines are free. What you pay for is the system around them: orchestrating 23 sensors, routing, parallelism, deduplication, contextual risk scoring, tiered reports, release gates, audit evidence, and ongoing updates — maintained for you. The AISG core (engine, correlation, policy, decision logic) is proprietary; sensors are open-source with full attribution. Like paying for a finished car, not loose engine parts. Building this yourself takes weeks, and you'd maintain it forever.

Can I trust results from CLI / Community-Edition tools?

Yes — these are the same engines used across the industry (many are OWASP/CNCF standard), and being open-source means they can be audited. What raw tools lack is the system: AISG normalizes, deduplicates, scores risk, and supports deterministic re-scans (same input → same findings, verifiable). Every finding records its tool, version, and context (provenance). Results are indicative and meant for human review — as stated in Trust & Legal.

Do I need to give you my API keys?

No. API keys are optional and only used for red-teaming paid LLM endpoints — and even then they are yours (BYOK), encrypted at rest, and never shown again. Free/local models (Ollama, vLLM) need no key at all. Without a key, the LLM layer is skipped transparently — everything else still scans.

Does AISG get access to my code?

No. Local scans run in-place where your code lives — code never leaves your machine. AISG only receives the findings (what was found, where, how severe), never your source code or keys.

What happens when my quota runs out?

You stay on your tier — nothing is downgraded. You simply can't start new scans until you top up (Scan Pack) or upgrade. AISG tells you exactly where you stand and what to do next, right on your dashboard.

What's the real difference between tiers?

Depth of report. Community shows findings count only (plus a free deep first scan). Beginner unlocks detailed findings + risk score + PDF report. Expert adds release gate, checklist, trends. Enterprise is the audit-grade layer: immutable audit trail, signed reports (SHA256), and the full audit report export (Markdown + PDF). Lihat contoh report dari pemindaian nyata.