Redloop 日本語 Book a scoping call

Continuous AI penetration testing for enterprises

Attackers test you every day. Your pentest runs once a year.

Redloop runs hard, sustained, authorized red-team testing across your external attack surface, cloud, identity, applications and people, using the latest AI models with senior operators in control. We find the way in before an attacker does, prove it, and retest the day you ship the fix.

Book a scoping call Get the 1-page brief

Every test is contractually authorized and scoped in writing. Nothing outside your signed scope is ever touched.

Attack path · examplechained
  1. Forgotten staging subdomain still answers on the internetexternalmedium
  2. Its config file exposes a long-lived CI tokenapplicationmedium
  3. The token can assume an over-permissive cloud roleidentitycloudmedium
  4. That role can read the customer-data bucketcritical path
Three “mediums” on three different reports. One way in. Redloop tests the chain, not the checklist.
Illustrative example of the kind of path we look for. Not client data.
20%

of breaches began with exploitation of a vulnerability, up 34% in a year.1

30%

of breaches involved a third party, double the year before.1

$10M+

average cost of a data breach in the U.S.3

The problem

A once-a-year pentest is a photo of a building that changes every day.

Sound familiar?

The report is out of date before anyone finishes reading it.New services, cloud accounts and SaaS logins ship every sprint.

Every finding says “medium”, and nobody can say which ones chain together.Lists of CVEs don’t show the path an attacker would take.

The board asks whether you’d stop a real attack. One PDF a year can’t answer that.Auditors want evidence of testing. Leadership wants proof it works.

Before The annual pentest

  • One window a year, one narrow scope
  • Findings arrive weeks later, in a PDF
  • Retesting costs extra, so fixes go unverified
  • Cloud, identity and people are tested by different vendors, or not at all
  • Everything that changes after the test goes unseen

After Redloop

  • Testing that never stops, across every surface you authorize
  • Each finding lands in your ticketing tool once it’s proven
  • Retest on every fix, included
  • External, cloud, identity, apps and social engineering in one program
  • New assets in scope get attacked as soon as they appear

What you get

Pressure like a real adversary. Paperwork like a real vendor.

01

Hard, sustained pressure

AI agents run attacker playbooks around the clock: recon, credential attacks, misconfiguration hunting, exploitation. Senior human operators approve every high-impact step.

02

Chains, not checklists

We connect low and medium findings into the paths an attacker would actually take, and rank them by what they reach: customer data, payments, production, domain admin.

03

Proof, fix, retest

Every finding comes with evidence, reproduction steps and a specific fix. When you ship it, we retest it automatically and close the loop.

Coverage

Every way in. One program.

You choose the surfaces. We test the ones you sign off, and nothing else.

External attack surface

Domains, IP ranges, VPNs and edge devices, exposed services, forgotten hosts.

Cloud

AWS, Azure and Google Cloud configuration, IAM paths, storage, secrets, Kubernetes.

Identity

SSO and MFA, Active Directory and Entra ID, privileged access, token and session abuse.

Applications & APIs

Web apps, APIs and business logic, authentication flows, LLM features you ship.

Social engineering

Phishing, phone and help-desk pretexting simulations, run only on staff lists approved by HR and legal.2

Authorized. Scoped. In writing.

We attack you because you hired us to, and only where you said we could.

Offensive testing without clear authorization is a liability for everyone. Every Redloop engagement runs under a signed contract and written rules of engagement before a single packet is sent.

No signature,
no test
  1. 01
    Written authorization firstAn authorized officer of your company signs the authorization and rules of engagement. Testing can’t start before that.
  2. 02
    An asset list you own or controlWe test only the domains, IP ranges, cloud accounts, apps and identities listed in the signed scope. Anything not on the list is out of bounds.
  3. 03
    Third parties stay out unless they agreeSystems hosted by vendors or SaaS providers are tested only where that provider’s published policy or written permission allows it.
  4. 04
    Windows, limits and exclusionsYou set testing hours, rate limits, fragile systems to avoid, and which techniques need extra sign-off.
  5. 05
    A named contact and a stop buttonYour team can pause or stop all activity at any time. Every action the AI agents and operators take is logged and shared with you.
  6. 06
    Proof, not plunderWe collect the minimum evidence needed to prove impact, never bulk data, and delete it at the end of the engagement as the contract specifies.

How it works

From scoping call to always-on in weeks, not quarters.

  1. Scoping call30 minutes on your surfaces, constraints and what “breached” would mean for you.
  2. Sign the scopeContract, asset list, rules of engagement and emergency contacts, in writing.
  3. Baseline attackA full first pass across every authorized surface, with a readout for your team.
  4. The loopContinuous testing as things change. Findings are filed in your tracker, proven and prioritized.
  5. Fix and retestShip the fix, we confirm it holds. Board-ready summaries every quarter.

AI with a short leash

The latest AI models do the grinding. Experienced people make the calls.

What the AI agents do

  • Map and re-map every authorized asset as it changes
  • Try far more attack variations than a human team has time for
  • Read your configs, code paths and permissions like an attacker would
  • Chain small weaknesses into realistic attack paths

What our operators do

  • Approve every high-impact or potentially disruptive action
  • Verify each finding before it reaches you (no raw scanner noise)
  • Design social engineering scenarios with your HR and legal teams
  • Brief your security team and leadership, in plain language

Engagement type

Pick how hard, and how often, you want to be tested.

One-off assessment

A deep first look, or a pre-audit check.

from$28,000 per engagement
  • 3-week engagement, up to 3 surfaces
  • Attack-path report with proof and fixes
  • One retest within 90 days
  • Executive readout

Quarterly

Regular pressure that keeps pace with releases.

from$19,000 per quarter, 12-month term
  • 4 assessments a year, all 5 surfaces in rotation
  • Unlimited retests on fixes
  • One social engineering campaign a year
  • Quarterly board-ready summary
Most coverage

Continuous

Always-on red teaming for every authorized surface.

from$14,000 per month, billed annually
  • Testing never stops across all 5 surfaces
  • Findings pushed to Jira, ServiceNow or Slack as they’re proven
  • Retest on every fix, automatically
  • Quarterly social engineering, named lead operator

Prices in USD, excluding tax. Final pricing depends on the size of your scope and is confirmed in writing after the scoping call. No tool to install on day one.

Founding engagements

We’re onboarding the first enterprise engagements in waves. Founding clients get first choice of start dates and founding pricing for their first 12 months. Reserving costs nothing.

Book a scoping call.

Three quick questions so the first call is useful. We’re onboarding in waves and will email you to schedule when your slot is ready.

  • 30 minutes with an operator, not a sales script
  • You leave with a draft scope and a price range
  • No testing of any kind before a signed contract
Scope (pick any)
Engagement type

No card, no commitment. We never test anything without a signed contract.

One page · PDF

How AI pentesting works, on one page.

The loop, the guardrails, where AI helps and where humans still decide, plus ten questions to ask any AI pentest vendor (us included). Easy to forward to your CISO or procurement.

Questions

The questions your team will ask.

Is this just a vulnerability scanner?

No. Scanners list known issues. Redloop tries to break in the way an attacker would: it chains weaknesses across surfaces, proves impact with evidence, and verifies every finding with a human operator before it reaches you.

Will testing disrupt production?

You set testing windows, rate limits and exclusions in the rules of engagement. Potentially disruptive techniques need explicit sign-off, operators approve high-impact steps, and your team can pause everything at any time.

Do you test our vendors or SaaS providers?

Only what you own or control, and third-party-hosted systems only where the provider’s published policy or written permission allows it. We never test systems outside your signed scope.

What if the AI does something it shouldn’t?

Agents work inside a technical allow-list built from your signed scope, every action is logged and visible to you, and operators approve anything high-impact. If something looks wrong, the stop button halts all activity.

How do social engineering simulations work?

Scenarios are agreed with your security, HR and legal teams, run only against an approved staff list, and are designed to teach, not shame. We don’t keep any real credentials that are entered.

Is Redloop available today?

We’re onboarding the first enterprise engagements in waves. Book a scoping call or reserve a founding spot and we’ll email you when your slot is ready. Nothing is charged until you sign a contract.

Book a scoping call

Reserve a founding engagement

Reservations open soon. Leave your work email and you’ll be on the priority list: first choice of start dates and founding pricing for 12 months. No payment details, nothing to sign today.