We test your AI agents the way an attacker would.

Parallax is a security research lab. We show your team exactly what we found, help you fix it, and confirm the fix works.

Request an assessment

The risk

AI agents don't just answer questions. They take actions.

They send email, update records and work in your other systems, using real access to your data. If someone hides instructions in something an agent reads, like a web page, a document or a support ticket, the agent may follow them.

The attack

We attack your agents the way a real adversary would.

  1. Prompt injection

    Hidden instructions in content the agent reads, such as a web page or an email.

  2. Tool abuse

    Tricking the agent into misusing the tools it is allowed to use.

  3. Connector hijack

    Using the agent's connections to your other systems to reach data it shouldn't.

The fix

Then we help your team close every step.

We walk your engineers through each issue and how to fix it. Your people make the changes, with us alongside.

The retest

We run the attack again to prove the fix holds.

Once your team has shipped the fixes, we repeat every step. Each one now stops at its gate.

The outcome

Launch AI agents with confidence.

Your business gets the AI agents it wants. Your security team knows exactly where the risks are, and has proof the issues are fixed.

Our team has served

What we do

AI agents are our focus. They are not all we test.

The same senior engineers test the systems around your agents, run full attack simulations, and help your team fix what we find.

Our focus

AI agent red teaming

We attack your AI agents the way a real adversary would, through everything they read, use and connect to.

  • Prompt injection

    Hidden instructions in content the agent reads, such as a web page or an email.

  • Tool abuse

    Tricking the agent into misusing the tools it is allowed to use.

  • Connector hijack

    Using the agent's connections to your other systems to reach data it shouldn't.

Around your agents

Penetration testing

We test the applications and infrastructure your business runs on, and show how far each weakness goes.

  • Web apps and APIs

    The applications your customers and staff use, and the interfaces behind them.

  • Cloud

    Your cloud accounts, and what each identity in them can reach.

  • Networks

    What an attacker can reach from outside, and from inside once they are in.

The whole organization

Red team engagements

A full attack simulation with goals we agree with you first. It shows how far a real attacker could get, and what your team would see along the way.

  • Goals you choose

    You decide what we go after, such as a system or a set of data.

  • One attack, start to finish

    We chain weaknesses across your systems the way an attacker would.

  • A debrief with your team

    We walk your defenders through every step we took.

After every test

Fix and retest support

We work with your own engineers to close what we found. Your people make the fixes, with us alongside.

  • A walkthrough of every issue

    We show your engineers what we did and why it worked.

  • A recommended fix for each one

    Written for the people who will make the change.

  • A retest

    We run the attack again to confirm the issue is closed.

How it works

Four steps, working alongside your team.

  1. Scope

    We meet with your team to agree what we test: the agents, applications and data in scope.

  2. Test

    We attack them the way a real adversary would.

  3. Fix

    We walk your engineers through each issue and how to fix it.

  4. Retest

    We test every fix and give you a final report.

What you get

A clear report, and fixes that hold.

  • Findings

    Every issue we found, explained in plain language.

  • Steps to reproduce

    Step-by-step instructions to reproduce each one.

  • Fixes

    A recommended fix for each issue, written for your engineers.

  • Retest

    A retest after you fix it, to confirm the issue is closed.

Who we are

A security research lab, led by senior engineers.

We are senior and principal offensive security engineers. We have brought our experience in advisory, AI and cybersecurity together to test AI agents, and the systems around them, for enterprises.

  • Where we come from

    Financial services, gaming, crypto and infrastructure.

  • Who we work with

    In-house security teams at enterprises, with a focus on financial services.

Founders

  • Travon Duncan
  • Tayvion Payton
  • Parker Lee

Tell us what you want tested.

Your AI agents, your applications or your whole organization. A short note is enough to get started.