NexGuard by LigeroNex

Prove what your AI agent can be made to do.

NexGuard attacks your AI agents in a sealed sandbox, shows the evidence for every finding and proves the fix holds, so your security team can sign them off. Everything below is a working preview with fictional data.

Try the demo ↓ How it works
  • Interactive preview
  • Fictional data
  • No sign-up needed

The problem

AI agents are stuck at security sign-off.

An agent that can read records, send email or move money is more than a chatbot. Security teams are asked to approve it with no repeatable way to show what it can be made to do.

Agents act, not just answer

They call tools, read records and take actions. A failure is something done, not a wrong sentence.

Ordinary tests miss it

Application security tests and model benchmarks do not cover how a model, its prompts and its tools behave together.

Every change reopens the question

A new model version, prompt, tool or permission can undo last month's approval.

How it works

From blocked to signed off, in six steps.

  1. 01

    Authorise

    Testing starts with written permission and an agreed scope. No permission, no test.

  2. 02

    Map

    List each agent's tools, permissions and the data it can reach.

  3. 03

    Attack safely

    Run versioned attack scenarios in a sealed sandbox, each several times.

  4. 04

    Show the evidence

    Every finding links to the exact steps, versions and run that produced it.

  5. 05

    Fix

    Your team tightens the tool, prompt or permission. People decide the severity.

  6. 06

    Retest and sign off

    Re-run to prove the fix holds, then give the approver one evidence pack.

Try it yourself

Take one agent from blocked to signed off.

Press Run the test set, open a finding, record the fix and retest. It takes about two minutes. The company, the people and the results are fictional.

This window is interactive. Click the menu, the cards and the buttons.

How it is designed

Built for the person who has to say yes.

NexGuard is designed around four ideas.

Evidence an approver can use

Scope, versions tested, every step of a failure and the history of its fix, in one pack.

The same answer every time

The first release uses four fixed checks over the recorded steps, not an AI opinion, so a result can be repeated.

Tests chosen for your agent

Scenarios are picked from what the agent can actually reach, and what was not covered is stated openly.

Contained by design

Attacks run only in an isolated sandbox, with decoy data and mocked tools. Zero escapes is a release rule.

Safeguards

Hard limits, not settings.

A product that attacks AI agents has to be safe to point at your own. These limits cannot be switched off.

  • Permission firstNo test without written authorisation from the system owner and an agreed scope.
  • Sandbox onlyNo attack runs outside an isolated sandbox that is destroyed afterwards.
  • Nothing real happensPayments, messages and deletions are simulated with mocked tools.
  • Decoy dataTests use decoy and synthetic data, not real customer data.
  • Yours aloneFindings and attack detail are never shared between customers.
  • People decideSeverity, accepting a risk and sign-off are decisions for named people.

Working with us

What a pilot looks like.

Small on purpose: one kind of agent, one environment, one scenario set. What counts as success is agreed in writing before it starts.

01

How much it catches

The share of the agreed scenarios where a real weakness is detected.

Measured against the agreed scenario set
02

How few false alarms

The rate of findings your own team judges not to be real problems.

Judged by your security team
03

How fast it gets fixed

The time from a finding to a fix that is proven by retest.

Compared with your process today

Where NexGuard is today. NexGuard is in development and we are looking for a first design partner. This site is a preview of the intended product, and every figure in it is illustrative. Whichever way a pilot goes, you keep the evidence pack, the results and a written list of what was not covered.

Next step

Have an agent stuck at sign-off?

Tell us about it. We will walk you through the preview and what a pilot could look like.

Talk to LigeroNex → info@ligeronex.com