Skip to main content
Yes AI - AI Solutions Australia
Referral service: an independent specialist firm does the testing

Security and Penetration Testing: Choosing the Right Kind of Test

Yes AI does not carry out security or penetration testing. Testing live systems needs professional indemnity and cyber insurance written for that work, and we do not hold it. When a business asks us for a test, we introduce it to an independent specialist security firm, which scopes the test, contracts with the business directly, invoices it directly and carries its own insurance.

This page describes in general terms the main kinds of security testing (external network, internal network, web application, cloud, social engineering, and testing of AI applications and chatbots), how a test is scoped, what to settle with a firm before it starts and how our introduction works. It is general information. Which tests suit your business, and what any results mean, are questions for the firm you engage.

What the Introduction Includes

6 test types
Explained on this page in plain terms
External, internal, web application, cloud, social engineering, and AI applications and chatbots
1 warm introduction
To an independent specialist security firm
By email or a short call, with only the context you have agreed we may pass on
4 checks
We make before introducing a firm
Its own insurance, qualified testers, a sample report before you commit, and a direct contract with you
0 testing
Done, supervised or resold by Yes AI
The firm runs the test and reports to you. We give no security opinions or warranties

Four Decisions That Shape Any Security Test

Security testing covers very different services sold under similar names. Settling these four points early makes quotes comparable and the final report useful.

Name the target before the method

A penetration test is defined by what is being tested. The addresses your business exposes to the internet, the office network, a customer portal, a cloud account, your staff and an AI chatbot each call for different skills, tools and access. A request for "a pen test" with no target cannot be quoted sensibly, and two firms will quote for two different jobs. Describe what the system does and what worries you, and let the firm propose the target and the method.

Decide how much the testers are told

Some tests start with almost no inside information, so the testers see what an outsider would see. Others give the testers user accounts, network diagrams, documentation or source code so they can cover more ground in the same number of days. Firms often call these black box, grey box and white box testing. Each answers a slightly different question, and a good firm explains the trade-off for your situation before you sign.

A test is limited by time

A penetration test runs for a limited time against an agreed scope. The report describes what the testers found in that time with that access. A short test with few findings does not show that no weaknesses exist, and a longer test against a larger scope is not automatically better value. Ask each firm how many days it proposes, what it will do with them and what it will leave out.

The firm carries the responsibility

Testing can disrupt a live system, and owners, customers and insurers rely on the results. The firm doing the work should hold professional indemnity and cyber insurance that covers it and should contract with you directly. Yes AI does not carry that cover, which is why we introduce a firm instead of taking the work on and passing it to someone else behind the scenes.

The Main Kinds of Security Testing

Firms use slightly different names, but most engagements fall into one or more of these six scope types. A single engagement can combine several of them.

Internet-facing view

External network testing

Testers look at what your business exposes to the internet: public IP addresses, remote access gateways, mail servers, firewalls and any services that answer from outside. They usually start with scanning to find what is reachable, then try to confirm and use any weaknesses within agreed limits. This kind of test is normally run remotely. Our network security scan page explains the difference between a scan and a test in more detail.

Inside-the-office view

Internal network testing

An internal test starts from inside the network, as if a laptop in the office had been compromised or someone had plugged a device into a spare wall port. Testers check what that starting point could reach: file shares, servers, printers, admin accounts and other devices. The firm may send a technician on site, or ship a small testing device or provide a remote connection that your IT provider sets up for the test.

Application logic tested

Web application testing

Testers work through a website, portal or API to see whether it allows things it should not, such as one customer viewing another customer’s records or a user reaching admin functions. Much of the effort is manual because these problems depend on how the application was built. Our website security audit page covers this kind of testing in more depth, including hosted platforms and test accounts.

Cloud configuration tested

Cloud testing

A cloud test looks at how your accounts on platforms such as Microsoft Azure, Amazon Web Services or Google Cloud are set up: who has which permissions, what storage is publicly readable, how services talk to each other and what is exposed to the internet. The major cloud providers publish rules on what customers may test. The firm should work within those rules and test only resources that belong to you.

How people respond

Social engineering testing

These tests check how staff respond to deception, for example a simulated phishing email, a phone call from someone claiming to be IT support or a visitor trying to get past reception. They need careful planning with management, and often with HR, about who knows in advance, how results are reported and how staff are treated afterwards. Agree with the firm whether results are reported by group or process, or by individual.

AI behaviour tested

AI applications and chatbots

A chatbot or AI assistant can be tested like any other application, plus some issues of its own. Testers try prompt injection, where crafted text in a message or document tries to make the AI ignore its instructions, and check whether it can be led into revealing data it should not, or into taking actions through connected tools. OWASP publishes a separate Top 10 for Large Language Model Applications, which lists prompt injection and sensitive information disclosure among its risks.

Situations Where Businesses Ask About Security Testing

TaskTraditionalThrough a Yes AI introductionNotes
A website chatbot answers customer questions from your documentsAssume the AI vendor has tested everythingIntroduced to a firm that can test your chatbot set-up and the data it can reachWhat the vendor has tested on its own platform is a question for the vendor. How your bot is configured and what it is connected to is a separate question for the firm.
An AI assistant can read the inbox, update the CRM or send messagesTest it the same way as a websiteFirm briefed on which tools and data the assistant can useConnected tools change what a successful prompt injection could do. The firm decides how to test that safely.
Moving systems into a cloud account for the first timeTrust the default settingsIntroduction to a firm that tests cloud configuration within the provider’s rulesAsk the firm to confirm which accounts and services are in scope and that it follows the provider’s published testing rules.
An office network that has grown device by device for yearsNobody knows what can reach whatFirm scopes an internal test; our systems register can list known devices and accountsThe register is an inventory you own. The firm decides what an internal test should cover.
Staff security training finished and the owner wants to know if it tookAsk staff whether they found it usefulIntroduced to a firm that runs planned social engineering testsAgree with the firm, and with HR if you have it, how results will be reported and used.
A customer-facing tool was built quickly with AI coding assistantsShip it because it works in the demoAn independent firm tests it before customers rely on itFast builds can still be tested properly. Tell the firm how it was built and who maintains it.
Two quotes for "a penetration test" differ widely in pricePick the cheaper oneCompare scope, tester days and sample reports with the questions on this pageDifferent prices often reflect different scope, method or days. The quotes need to describe the same job before price means anything.
A new release changes how customers log inRely on the original test from launchIntroduction to a firm to scope a focused test of the changed areasA test describes the system on the day it was tested. The firm decides whether a focused or full retest suits the change.

Common Problems With Security Testing Engagements

Testing an AI system with real customer data

Prompt injection and data leakage tests try to make a chatbot reveal what it can see. If the bot is connected to live customer records, a successful test can put real personal information into the testers’ notes and report. Ask the firm whether testing should run against a copy with test data, and how it will store and delete anything it sees.

Running social engineering tests without planning

An unplanned phishing or phone test can upset staff, damage trust and trigger real incident procedures. Decide in advance who in the business knows, what the testers may and may not do, how a staff member can check whether something is a test, and how results will be shared. A professional firm will raise these points itself.

Forgetting third parties that own part of the system

Your systems often sit on infrastructure owned by others: a cloud provider, a SaaS vendor, a hosting company or a managed IT provider. You can only authorise testing of what you control, and some providers publish rules that limit what customers may test. The firm should confirm these permissions with you before testing starts.

No agreed way to stop a test

Every test should have named contacts on both sides who can be reached during testing and who can pause it if something slows down or breaks. Agree the testing hours, the contacts and what counts as a reason to stop. Tell whoever monitors your systems, so a test is not mistaken for a real attack, unless you and the firm have deliberately agreed to see whether monitoring notices it.

Reading a short report as a clean bill of health

A report with few findings may mean the system held up well, or that the scope, days or access were narrow. Read the scope and limitations section as carefully as the findings, and ask the firm what it would test next with more time. Yes AI does not interpret reports. That conversation is between you and the firm.

The same people testing and fixing

If the team that built or runs a system also tests it, it is marking its own work. Many businesses prefer testing by a firm with no stake in the build, while their developer or IT provider makes the changes. Decide which you want before you ask for quotes, so the firm can tell you whether it offers remediation help and how it keeps that separate.

What Yes AI Does Next to the Introduction

A warm introduction

After a short call we introduce you to an independent specialist security firm by email or a three-way call, passing on only what you agree to share. Before introducing a firm we check that it holds its own professional indemnity and cyber insurance, that its testers hold recognised security qualifications, that it will show you a sample report before you commit, and that it contracts with you directly.

Documentation for AI systems we built

If we built your chatbot or AI agent, we can give the firm the documentation you approve: what data the system can reach, which tools it can call and how it is hosted. That saves discovery time. We do not test the system ourselves, and the firm decides what to test.

A register of your systems

Our systems audits produce a register you own of accounts, admins, licences, domains, phone numbers and SaaS tools. If you choose to share it with the firm, it can help with scoping. It is an inventory and does not assess security.

Development work carries on

If the firm reports findings about something we built for you, changes to it are ordinary development work you can ask us to do, and the firm confirms any fix when it retests. If we receive a referral fee from the firm, we tell you before the introduction, and it never changes what you pay the firm.

How the Introduction Works

Five steps. Yes AI is involved in the first three. The test itself is between you and the firm.

Book a call and describe what you want tested

Tell us what the system does, who built it, where it runs and what prompted the question, such as a launch, a customer request or a new AI tool. The initial consultation is free for businesses with 20 or more full-time staff, otherwise $200 including GST.

Agree what we may pass on

You decide what context the firm receives from us: the systems involved, any deadline, your developer or IT provider’s details and, if you wish, documentation or a systems register from work we have done for you.

Warm introduction

We introduce you and the firm. From that point the firm deals with you directly. If a referral fee applies, we tell you before this step.

The firm scopes and quotes

The firm proposes the test type, scope, tester days, timing, rules of engagement and price, and asks for written authorisation. You decide whether to go ahead. Yes AI is not a party to that contract.

Testing, report and retest

The firm runs the test, reports to you and answers your questions about findings. Any AI, automation or integration work you have with us continues as normal.

FAQ

Do you do the security work yourselves?

No. Security testing needs professional indemnity and cyber insurance written for that work, which the independent specialist firm carries and Yes AI does not. We are an introducer only. We do not perform, supervise, resell, white-label or subcontract security testing, and we do not give security advice, opinions or warranties.

Who should not use this introduction?

Anyone dealing with an active incident, such as ransomware, a takeover of staff accounts or a chatbot that is already leaking data, should not wait for an introduction. Contact your cyber insurer’s incident line if you have a policy, and report to the Australian Signals Directorate’s Australian Cyber Security Centre through cyber.gov.au, which also lists its 24 hour hotline. Introductions suit planned testing.

What does testing an AI chatbot involve?

Testers treat the chatbot as an application with an unusual input: natural language. They check the usual things, such as who can reach the bot and what data it can see, and then try to make it misbehave, for example by ignoring its instructions, revealing data from its documents or other users, or using connected tools in ways it should not. The firm agrees the method and the test data with you first.

What is prompt injection?

Prompt injection is when text supplied to an AI system, either typed by a user or hidden in a document, email or web page the system reads, is written to override the instructions the system was given. It is listed in the OWASP Top 10 for Large Language Model Applications. How much harm it can do depends on what the AI can see and do, which is why testers ask what data and tools it is connected to.

What are black box, grey box and white box tests?

They describe how much the testers are told. In a black box test they start with little or no inside information. In a grey box test they get some, such as user accounts or a list of systems. In a white box test they may see documentation, configuration or source code. More information usually lets testers cover more in the same time. The firm recommends an approach for your scope.

Can a security test break our systems?

Testing can affect live systems, for example by sending heavy traffic, filling in forms many times or locking accounts. Professional firms reduce this with agreed testing hours, limits on risky techniques, named contacts who can stop the test and, where sensible, testing against a copy of the system. Ask the firm how it handles this before you sign. Its insurance should cover the work it does.

Do we have to use the firm you introduce, and how is Yes AI paid?

No. You are free to decline the firm’s proposal or find another firm using the questions on this page. The booking goes through our normal call booking, the firm quotes and invoices its own work, and we publish no security prices. If we receive a referral fee from the firm, we tell you before the introduction, and it never changes what you pay the firm.

Ask for an Introduction to a Security Testing Firm

Book a call and tell us what you want tested, including any AI chatbot or assistant. We introduce you to an independent firm that carries its own insurance, and it scopes and quotes the test with you directly.

All discussions held in confidence. Australian-based consultants.

Interactive controls are loading. Phone and email links are available.