Pentest for startups: what to test at each stage, and when to pay for one

A startup needs to know what a stranger can do to its app from the first deploy, and that costs little. It needs a manual pentest later: when a customer contract, an auditor or card rules ask for a report signed by a tester. Until then, close the obvious leaks before launch, attack the app again on every deploy, and keep dated proof you can show a buyer.

The team behind Keybrum's attacksUpdated 10 min read

In short

  • Nobody audits you on day one. The first pentest request usually comes from a customer's security questionnaire, and the answer they want is a dated report and a list of what is still open.
  • SOC 2 has no explicit pentest requirement, and most apps that take cards through a hosted checkout fall outside the PCI DSS one. A customer contract that asks for a pentest makes it required anyway.
  • Fix the cheap findings before you pay a tester. A manual test that spends its days on open tables and leaked keys is the most expensive way to find them.
  • Scope the first manual test on the main app and its API, with two accounts per role, one retest and a letter you can forward.
  • A pentest describes one build. If you ship every week, keep an automated attack on every deploy between two human tests.
On this page

Does a startup need a pentest?

Yes, but probably not the one you are picturing, and not yet. A pentest for a startup is two different purchases that share a name. The first is knowledge: what can a stranger do to my app today? You need that from the first deploy, and you can get most of it for free or for the price of a lunch. The second is a document: a report by a named tester that a customer or an auditor will accept. You need that when someone asks for it, and it costs thousands.

Mixing the two is the common mistake. Founders either skip testing because "a pentest costs $10,000", or buy the $10,000 test before anyone asked and before the obvious holes were closed. This guide puts the two in order. For the numbers behind each option, see the penetration testing cost guide.

Why young apps get hit

Attackers do not pick a startup because it is famous. They find it because it is public, and scanning the public internet is cheap. Three things make a young app an easy find.

  • It is online from the first deploy. The database, the storage bucket and the API answer the whole internet on day one, long before the first customer.
  • It runs on defaults. A table created without row level security, a storage bucket left public, Firebase rules still in test mode. Nobody chose these settings. They were what the generator or the tutorial left behind.
  • It changes every day. With a coding agent, one prompt adds a table, a route and a migration. Each one is a new place for the same mistakes.

The good news hides in the same numbers. The most common flaws are also the cheapest to find, because a machine can prove them from outside. The vibe coding security checklist lists them with a live test for each.

Who will ask you for a pentest, and what they accept

The date of your first manual pentest is rarely your decision. Someone asks. Knowing who, and what they accept, saves you from buying the wrong thing.

Who asksWhat they ask forWhat the rule really says
An enterprise customer"Do you run penetration tests? When, by whom, and are high findings open?"Their own policy. Most want a third-party report from the last twelve months. See answering security questionnaires
A SOC 2 auditorEvidence that you evaluate your controlsNo explicit requirement. A CPA writing for the Cloud Security Alliance puts it this way: "there are no explicit requirements for penetration testing (or any controls) within a SOC 2 report". If a customer contract requires a pentest, it is required
Card rules (PCI DSS)A test at least once a year and after significant changesOnly when requirement 11.4 is in your scope. Most apps that send cards to a hosted checkout are outside it. See PCI DSS penetration testing
A security framework you adoptedA testing programCIS Control 18 asks for external penetration tests "no less than annually", by a qualified party, and only from its implementation groups 2 and 3, not from the smallest organisations
Your own team"Are we exposed?"No rule. This is the knowledge purchase, and an automated attack answers most of it

Two habits follow from this table. When a buyer asks for a pentest, ask them what they accept before you buy: a third-party manual report, a report from a testing platform, or dated evidence of automated tests. And never answer "yes, we are pentested" on the strength of a scanner. A gap stated honestly, with a date to close it, is a normal answer from a young company.

What to test at each stage

  1. 1

    Before launch: close what a stranger can prove

    Run the free checks on your live preview, not on your code. Tables without row level security, secret keys in the JavaScript bundle, .env files served by mistake, a second account that can open the first one's records. An hour of your time, and the Supabase RLS checker or the API key leak checker to start.

  2. 2

    First users: one full automated attack

    Before real personal data arrives, have the live app attacked once from outside, with two accounts, and fix what it proves. Keep the report and its date. It is your first piece of evidence.

  3. 3

    Paying customers, weekly deploys: attack every deploy

    The risk is now the change you ship on Friday. An automated test on every deploy catches the new table or the new route. A public page that shows what was tested and when starts answering buyers before they ask.

  4. 4

    First enterprise deal or audit: the manual pentest

    Now the document matters. Buy a manual test of the main app and its API, with a retest and a letter. Keep the automated attacks running between two tests, so the fixes stay fixed.

  5. 5

    Regulated data: a program

    With card data in scope, health data or a framework like CIS, testing becomes a calendar: at least yearly, after significant changes, with findings tracked to closure.

How to scope a first manual pentest

A manual test is priced in tester days, so scope is the price. A tight scope on what matters beats a wide one that is skimmed. For a SaaS app, this is the scope that usually satisfies a first enterprise buyer.

  • The production app and its API, or a staging copy that is truly identical, including the database rules. Tests on a staging app with different policies prove little.
  • Grey box: the tester gets accounts, so the days go to your access control and not to guessing passwords. The white box guide compares the three levels of access.
  • Two accounts per role and per tenant, so the tester can check that one customer cannot reach another customer's data. For SaaS penetration testing this is the question buyers care about most.
  • The API on its own line, with the list of endpoints. See API penetration testing.
  • One retest included, with a deadline you can meet. Fixing ten findings takes longer than you think.
  • The deliverables in writing: the full report, a summary without technical detail that you can send to customers, and an attestation letter with dates and scope.
  • What is out of scope, stated in the report: third-party services, social engineering, denial of service.

Ask each vendor for a sample report before you sign, and read one finding in full. A finding you can act on shows the request, the response, the account used and the fix. The pentest report sample shows what to look for, and the methodology guide explains the phases a tester should follow.

What to do with the report

A report in a drawer answers no questionnaire. Four steps turn it into something a buyer can read.

  1. Fix by proof, not by severity label. Start with what gives a stranger data. A critical label on a missing header can wait behind a medium that leaks invoices.
  2. Retest and keep the result. A finding is closed when the same request fails, with a date. "We fixed it" is not evidence.
  3. Write the short version. One page: what was tested, when, by whom, how many findings by severity, how many are closed. This is what you attach to a questionnaire. Never send the full report without a non-disclosure agreement.
  4. Schedule the next one. Put the next manual test on the calendar, and decide what runs between the two.

Budget: what each stage really costs

The first two stages cost almost nothing: free checks, then one automated report. Keybrum charges €49 for that report, once. Continuous testing on every deploy is a subscription: €29 a month for five apps or €99 a month for twenty-five at Keybrum, excluding VAT.

The manual test is the large line. Public list prices for one small web app start near $5,000, and a quote grows with roles, endpoints and the report format an auditor wants. The cost guide lists published prices by vendor and the day rates behind them. Two things keep the quote down: a clean scope, and an app whose easy findings are already closed, so the tester's days go to what only a person can find.

Five mistakes that waste a startup's pentest money

  • Buying the manual test first. The tester spends day one reporting open tables and a leaked key. You paid a day rate for what a free query finds.
  • Testing a staging app that differs from production. The policies, the keys and the data are not the same. The report describes an app your customers do not use.
  • Accepting a scanner report labelled pentest. If no finding shows a request that worked, it is a list of maybes. See pentest vs vulnerability scan.
  • One test, then nothing for a year. A team that deploys weekly has shipped fifty builds the report never saw.
  • No retest. Without it you hold a list of problems and no proof that any is closed.

A pentest checklist for startups

  • Row level security or database rules on every table and bucket, checked on the live app
  • No secret key in the JavaScript the browser downloads
  • Two accounts you own cannot read or change each other's records
  • AI routes and webhooks refuse callers with no session or no signature
  • A dated report of the last full attack, automated or manual
  • A retest result for every finding you say is fixed
  • An automated test on every deploy
  • A one-page summary and, when a buyer asks, a third-party manual report less than twelve months old
  • A written answer to "what do we accept as a pentest?" from each customer who asked

Where Keybrum fits, and where it stops

Keybrum is the automated pentest we build for apps made with AI. It covers the first three stages of this guide: it attacks the live app as a stranger and as two accounts, keeps the request that worked for every finding, hands the fix to your coding agent, and attacks again after the fix and on every deploy. Plans add a public trust page that shows what was tested and when. It only attacks an app after its owner proves ownership.

It does not replace the document of stage four. When an auditor or an enterprise customer asks for a report by a human tester, buy that test, and ask them first what they accept. Keybrum's job at that point is to make the tester's days count, and to tell you whether the fixes held the week after.

Questions

When should a startup get its first pentest?

Test what a stranger can do before launch, with free checks and one automated attack. Buy the first manual pentest when a customer contract, an auditor or a regulation asks for a report, which for most SaaS startups is the first enterprise deal. Buying it earlier is rarely wrong, but fix the obvious findings first.

How much does penetration testing for a SaaS startup cost?

An automated report costs from €49 at Keybrum, and continuous testing from €29 a month excluding VAT. A manual test of one small web app starts near $5,000 at public list prices and grows with roles, endpoints and the report an auditor wants. The penetration testing cost guide has the published prices.

Does SOC 2 require a penetration test?

Not explicitly. The SOC 2 criteria do not list a pentest as a required control, and auditors accept other evidence that you evaluate your controls. In practice many auditors and most enterprise customers expect one, and a contract that asks for a pentest makes it required.

Is an automated pentest enough for a startup?

It is enough to know whether a stranger can read your data, and to keep that answer current on every deploy. It is not enough when a buyer or an auditor asks for a report by a human tester, or when your risk is in business logic. See automated penetration testing for what a machine can and cannot prove.

What is SaaS penetration testing?

It is a pentest scoped for a multi-tenant web app: the application, its API and above all the wall between tenants. The central question is whether one customer's users can reach another customer's data, so the test needs two accounts in two tenants. See IDOR and broken access control.

How often should a startup pentest?

Follow whoever asks. CIS Control 18 and most enterprise buyers expect an external test at least once a year, and PCI DSS adds a test after significant changes when it applies. Between two manual tests, an automated attack on every deploy covers the builds the yearly test never sees.

Sources

  1. 1.SOC 2 Reports and Penetration Tests, Cloud Security Alliance
  2. 2.CIS Control 18: Penetration Testing, Center for Internet Security
  3. 3.Information Supplement: Penetration Testing Guidance, v1.1, PCI Security Standards Council
  4. 4.Statement on CVE-2025-48757, Matt Palmer
  5. 5.CVE-2025-48757, NIST National Vulnerability Database

Your app is public. So are its mistakes.

Check what any attacker can already see in it, free, in ten seconds. The full attack, with proof and fixes: €49 once, or €29 a month on a plan.

or continue with Google or GitHub