Subodh KcSubodh KcAI Advisor
Discuss AI
AI APPSEC · OPEN SOURCE · POWERED BY HAIEC

Evidence-backed AppSec for AI applications and agents.

Audit AI/LLM application code before commit, PR, merge or deployment.

AI AppSec combines a pinned static-analysis engine with explicit coverage, structured findings, scan receipts and proof-of-fix so the result says both what was found and what was actually checked.

v0.1.0MIT122 detectors79 security checksMCP v2Local scanning

The question it answers

Is this AI application code introducing a security risk?

AI AppSec is intentionally focused on AI/LLM application source security. It checks for security signals across the code developers are shipping, rather than attempting to provide complete assurance for the entire AI system.

What v0.1.0 checks

AI AppSec checks for security risks and signals including:

Unsafe AI-output execution

Secrets exposure

RAG and model-integration risk

Insecure AI API usage

Security control gaps

Prompt/input-related risk signals

These are security signals, not guarantees of detection. Finding kinds include PRESENCE, RISK_SIGNAL, CONTROL_GAP, and VULNERABILITY with severities from CRITICAL to INFO.

How the scan works

Scan pipeline
1TARGET
2SCOPE + FILE DISCOVERY
3SUPPORTED FILE ACCOUNTING
4SEMGREP 1.173.0 + BUNDLED PUBLIC CORE
5122 DETECTORS
679 SECURITY CHECKS
7FINDING INSTANCES
8CONCERN FAMILIES
9COMPLETE / PARTIAL / ERROR
10SCAN RECEIPT + EVIDENCE ENVELOPE
11PROOF-OF-FIX RESCAN

AI AppSec uses a pinned Semgrep 1.173.0 execution engine with a bundled Public Core rulepack. The package adds AI-focused security checks, explicit coverage accounting, evidence semantics, scan receipts, proof-of-fix and the MCP interface around that execution. Semgrep is an external engine installed separately. AI AppSec does not claim to have invented Semgrep.

The scan reports what it did not prove.

COMPLETE

All supported files in the intended scope were analyzed.

PARTIAL

Some intended evidence could not be analyzed. Absence of findings is not treated as proof.

ERROR

The scan itself could not complete.

Missing evidence is not a pass. A PARTIAL scan cannot establish absence of findings.

Coverage accounting includes:

DISCOVEREDINTENTIONALLY_EXCLUDEDUNSUPPORTEDTARGETEDENGINE_REPORTED_SCANNEDPARSE_FAILEDSUCCESSFULLY_ANALYZED

Evidence, not just output.

Scan Receipt

Tamper-evident digest tying together the finding and coverage result. SHA-256 digests for finding set, concern family set, coverage file sets, and evaluated checks.

Evidence Envelope

Binds receipt information to execution and target context, including producer identity, target identity, execution status, and envelope digest.

Semantic Security Checks

Higher-level grouping of findings into concern families without pretending every group is one root cause. Underlying finding instances remain auditable.

Proof-of-Fix

Compare a rescan against prior evidence without requiring unrelated code to remain identical. Verifies that a specific security check's findings were resolved.

Use it with an AI coding agent.

AI AppSec exposes a focused security check through MCP. The coding agent can request a scan, inspect findings, explain findings, help change code, and rescan. The independently executable scanner determines the structured findings.

MCP configuration
{
  "mcpServers": {
    "ai-appsec": {
      "command": "npx",
      "args": ["ai-appsec"]
    }
  }
}

Primary MCP tool: scan_ai_security

Natural language examples your agent can handle:

  • "Audit this AI agent before I push."
  • "Review this change for security problems."
  • "Check this AI application before merge."
  • "Run a security scan on this code."

Install and setup

Install globally
npm install -g ai-appsec
Check prerequisites
ai-appsec doctor

Checks prerequisite status (Semgrep availability).

Install Semgrep engine
ai-appsec setup

Installs and configures the required Semgrep engine. May require network access. Normal scans operate locally after prerequisites are installed.

Semgrep is NOT bundled with this package. It is an external engine installed separately. See THIRD_PARTY_NOTICES.md in the repository for details.

Limitations

  • -Static analysis only - does not execute target code
  • -No runtime behavioral assurance
  • -No compliance certification
  • -No full-system AI assurance
  • -Semgrep must be installed separately (ai-appsec setup)
  • -Setup may require network access
  • -Tenant isolation is a separate product (MCP Tenant Isolation)
  • -Runtime LLM verification is a separate product (LLMVerify)

Need tenant boundaries?

MCP Tenant Isolation

Need runtime/model interaction guardrails?

LLMVerify

Technical identity

ProductAI AppSec
Versionv0.1.0
GitHubsubodhkc/ai-appsec
npmai-appsec
MCP Registryio.github.subodhkc/ai-appsec
CLIai-appsec
MCP toolscan_ai_security
LicenseMIT
Node>=22
MCPv2 server (stdio)
BrandPowered by HAIEC

HAIEC DEVELOPER SECURITY

SOURCE SECURITY

AI AppSec

Current page

BOUNDARY SECURITY

MCP Tenant Isolation

Explore ->

RUNTIME

LLMVerify

Explore ->

Secure the code. Protect the tenant boundary. Verify the model interaction.

MIT licensed. Powered by HAIEC. No HAIEC account or API key required.

AI Advisor →