Is AI-generated code safe to deploy?

AI-generated code frequently contains security vulnerabilities that are invisible without specialized scanning. Common issues include hardcoded API keys and credentials, OWASP Top 10 vulnerabilities like injection flaws, insecure dependencies, backdoors introduced through prompt injection, and supply-chain risks in AI-recommended packages. VettIQ scans for all of these before code reaches production.

How do I scan code written by Cursor or GitHub Copilot for vulnerabilities?

VettIQ integrates directly with any AI coding environment. You connect your repository, and VettIQ runs a five-model sequential scan — detection, deep analysis, adversarial probing, fix suggestion, and verification — on all AI-generated code before it is merged. It works with Cursor, GitHub Copilot, Claude, and any other AI coding tool.

What is an MCP server and why does it need security scanning?

An MCP (Model Context Protocol) server extends AI agents with new capabilities — file access, API calls, browser control, and more. Connecting an insecure MCP server to an AI agent can expose your system to data exfiltration, unauthorized API access, and prompt injection attacks. VettIQ's MCP directory scores over 1,257 MCP servers for security risk before you connect them to your agents.

What is AI agent runtime security?

AI agent runtime security enforces guardrails on an AI agent while it is actively running — not just before deployment. VettIQ monitors agent behavior in real time, blocking unauthorized network calls, preventing data exfiltration, and stopping out-of-scope actions. This is critical for autonomous agents that operate without continuous human supervision.

How is VettIQ different from Snyk or traditional SAST tools?

Traditional SAST tools like Snyk are designed for human-written code with known vulnerability patterns. VettIQ is purpose-built for AI-generated code, which introduces novel risk categories — adversarial patterns from prompt injection, statistically plausible but insecure code constructs, and agent-level runtime behaviors — that traditional scanners were not designed to detect. VettIQ also covers the MCP ecosystem and agent runtime layer, which no traditional SAST tool addresses.

NemoClaw Secures the Runtime. VettIQ Secures the Supply Chain.

The Two Halves of Agent Security

When you deploy an AI agent, you face two distinct security problems:

Runtime security — what happens when the agent runs. Can it escape its sandbox? Can it access files it shouldn't? Can it make network calls to unauthorized endpoints?
Supply chain security — what goes into the agent environment. Are the MCP servers and skills the agent uses actually safe? Has anyone checked them for prompt injection, credential theft, or malware?

NemoClaw answers question #1 with OpenShell sandboxed execution. VettIQ answers question #2 with the MCP Trust Directory. Together, they provide defense in depth.

What NemoClaw Brings to the Table

NemoClaw (announced at NVIDIA GTC, March 16, 2026) is an enterprise-grade runtime security layer for AI agents. It provides:

OpenShell containers that isolate agent actions from the host system
Policy-based access control defining what the agent can and cannot do
Privacy routing that keeps sensitive data on local Nemotron models instead of cloud APIs
Action logging for compliance audit trails

This is exactly what the AI agent ecosystem needed. Before NemoClaw, most agents ran with zero runtime guardrails — full network access, no approval gates, no sandboxing.

The Gap: A Malicious Skill in a Sandbox Is Still Malicious

NemoClaw sandboxes execution. But it trusts whatever you install inside the sandbox.

Consider the MCP ecosystem. There are now over 2,000 MCP servers on GitHub. Developers install them into Claude Code, Cursor, Windsurf, and NemoClaw environments with a single command. Most have never been security-reviewed.

A sandboxed agent running a malicious MCP server can still:

Exfiltrate data through approved network endpoints
Inject malicious prompts that manipulate the agent's behavior
Abuse API keys that the agent legitimately has access to
Read sensitive files within the sandbox's mounted volumes

The sandbox contains the blast radius. But the damage still happens inside the container.

How VettIQ Fills the Gap

VettIQ operates upstream of NemoClaw. Before a skill enters the sandbox, VettIQ has already scanned it through four security engines:

Snyk — prompt injection and code vulnerability detection

Cisco Kenna — dependency risk scoring across the full tree

Semgrep — dangerous code patterns (eval, exec, credential access)

VirusTotal — malware signature matching against 70+ engines

The results are published to the MCP Trust Directory — a free, public database of 1,000+ MCP servers, each with a risk score from 0 to 100. Green (0-30) means safe. Amber (31-65) means use with guardrails. Red (66+) means don't install.

The Workflow

1. Developer finds an MCP server on GitHub

2. Checks vettiq.ai/mcp for trust score

3. Sees: APPROVED (risk score 12/100)

4. Installs into NemoClaw with confidence

5. NemoClaw enforces runtime guardrails

Result: defense in depth

Not a Partnership — a Complement

VettIQ is not a NemoClaw product or partner. We're an independent security layer that works alongside any agent platform — NemoClaw, OpenClaw, Claude Code, Cursor, Windsurf, or custom agents.

NemoClaw happens to be the strongest runtime complement because it takes sandboxing seriously. VettIQ takes supply chain vetting seriously. Together, they cover the full attack surface.

Get Started Today

VettIQ's MCP Trust Directory is live and free:

Browse 1,000+ vetted MCP servers — no login required
Download a Security Blueprint — free rules for AI coding tools, including the new NemoClaw + OpenShell blueprint
Read the NemoClaw use case — detailed breakdown of how VettIQ + NemoClaw work together