Penetration testing for SOC 2: requirements, scope and audit evidence
SOC 2 does not hinge on a pentest checkbox. It hinges on evidence: that you tested the right systems, fixed the findings that mattered, and can show the trail to an auditor. That is why teams usually fail on scope, timing or documentation, not on the test itself.
Key takeaways
- SOC 2 does not explicitly mandate a penetration test, but many programs still need one to satisfy the auditor and customer security reviews.
- For Type II, the useful evidence is the full chain: test, remediation and retest, all inside the observation period.
- Scope must match the SOC 2 system boundary: production apps, APIs, authenticated flows and the relevant external exposure.
- A scan alone is not enough when the control question is exploitability or business-logic abuse.
SOC 2 does not mandate a pentest, but you may still need one
What the Trust Services Criteria actually say
No Trust Services Criterion requires a penetration test by name. The criteria ask whether you identify weaknesses, evaluate your controls and respond to what you find. As Stingrai's analysis of the 2022 points of focus puts it, "no Trust Services Criterion names one as a mandatory control, and the 2022 points of focus list penetration testing only as one example of a CC4.1 separate evaluation."
CC4.1 is the monitoring criterion: the entity "selects, develops, and performs ongoing and/or separate evaluations to ascertain whether the components of internal control are present and functioning," as the Cloud Security Alliance quotes it. A pentest is one of the most convincing ways to perform that separate evaluation, not the only one.
Where the requirement usually comes from
In practice the need for a pentest comes from three places: your system description, the CPA firm's evidence expectations and customer security questionnaires. A SaaS company can cover some controls with vulnerability scans, then discover the audit file still lacks a deeper test of authenticated flows, admin paths or tenant isolation.
The question an auditor asks is narrower than most teams think. It is not "did you run a pentest?" It is whether you can show that you identify exploitable weaknesses, check that your controls work and resolve the findings that matter.
Type I, Type II and why timing matters
Point in time versus observation period
Type I is a point-in-time view. Type II covers operating effectiveness over an observation period, usually several months, so the question becomes whether the control worked during that window, not just whether it existed on day one.
| Timing question | Type I | Type II | What to do |
|---|---|---|---|
| When is evidence evaluated? | Point in time | Over an observation period | Match testing to the audit window |
| Must test and remediation fit the same window? | Not always | Yes, for operating effectiveness | Test early enough to fix and retest |
For Type II, testing, remediation and retesting should all fall inside the observation period. A pentest run after the period closes may still be a good security result, but it is weak SOC 2 evidence.
Schedule the test early
Run the main test in the first half of the window. On a six-month Type II period, landing it around month two or three leaves room for triage, fixes, a focused retest and evidence cleanup before the auditor asks for artifacts. A clean report with no retest inside the window is weaker than a narrower test that closes the loop.
Remediation speed is the usual constraint. According to Fluid Attacks' State of Attacks 2025, the mean time to remediate vulnerabilities was about 55 days. If your fix cycle is longer than the time left in your window, test earlier or tighten ownership of the remediation queue.
Scope the test to your system boundary
What belongs in scope
You are not testing "the company." You are testing the services, interfaces and environments that support the service you described to the auditor. For most SaaS teams that means:
- the production web application and its authenticated workflows;
- public and private APIs;
- cloud resources and external infrastructure that can affect the service;
- integration points with third-party identity, messaging or storage providers (you test your side of the integration, not the vendor).
What teams wrongly leave out
Admin panels, tenant-isolation logic, staging environments that touch production data and APIs inherited from fast shipping are the usual gaps. They do not look like customer features, but they often carry the easiest privilege-escalation paths. The login page is rarely the interesting part; what happens after login is.
Take a 20-person B2B SaaS with a React frontend, a GraphQL API, background jobs and an internal admin panel. The boundary should include login, role-based access checks, invitation and billing flows, tenant switching and every endpoint the app or the support team uses. Chained actions across those flows are where a pentest earns its keep.
A vulnerability scan is not enough
A scanner is automated and broad. It flags missing patches, weak headers and known CVEs, which supports vulnerability management. It cannot tell you whether a low-severity issue becomes an account takeover when chained with a broken access check.
A penetration test validates exploitability, usually through authenticated paths, chained requests or business-logic abuse. That is the difference between "something is misconfigured" and "this user can read another tenant's data." A scanner may flag an outdated library; a tester finds that a user can escalate privileges by combining the invite flow, a weak tenant check and a reused session token. We cover the distinction in more depth in vulnerability scanning vs penetration testing.
NIST SP 800-115 remains the common methodology reference: it structures planning, execution and reporting without imposing a tool or a vendor.
What an auditor-ready pentest package contains
Minimum contents
The report has to answer who tested what, when, how, what was found and how it was closed:
- Scope: in-scope systems, exclusions and boundary assumptions.
- Dates: test start and end, fix dates, retest dates.
- Method: authenticated testing, controlled exploitation, methodology reference.
- Findings: the issue, its impact and how it was reached.
- Severity: a clear priority so the auditor can follow the triage logic.
- Remediation: tickets, owners and resolution status.
- Retest: proof the issue was fixed, or why it remains open.
What remediation evidence looks like
Open findings are not fatal. If they are triaged, tracked and backed by a remediation plan or a compensating control, auditors care more about the process than about every issue being closed on day one. The problem is a missing workflow, not the presence of risk.
Map the evidence to CC4.1, CC7.1 and CC7.2 where it fits, but do not claim a pentest covers all of SOC 2. It supports control evaluation and vulnerability management. It does not replace access reviews, change management or incident response evidence. The same logic applies to other frameworks; see what ISO 27001 auditors expect from a pentest.
Running a SOC 2 pentest without breaking delivery
A risk-based cadence
Annual testing is common, but it should not be your only trigger. A new authentication flow, a new cloud region, a new public API or a change to tenant routing changes the attack surface. That is when to test again, whatever the calendar says.
The workflow
- Define the system boundary with engineering and compliance so the test matches the audit scope.
- Limit risky releases during the test window so fixes do not fight new deploys.
- Authorise the test in writing: exclusions, targets, test accounts.
- Triage findings fast and assign owners before the report goes stale.
- Push remediation into the tools the team already uses (Jira, Linear, Slack).
- Retest each fix and archive the retest note with the original evidence.
The trade-off is simple. A one-off annual test may satisfy a checkbox; testing at every significant change fits fast-moving products but needs someone who owns the remediation queue.
A checklist for your next audit cycle
Before testing
- Confirm with your CPA firm whether they expect a pentest or accept alternative evidence.
- Lock the system boundary so the test matches the SOC 2 description.
- Align the test date with the Type II observation window, not the audit deadline.
- Write down exclusions, including staging, third-party services and out-of-scope hosts.
- Decide in advance which severity triggers a mandatory retest.
After testing
- Store the report, remediation tickets, retest notes and management sign-off together.
- If you skip the pentest, agree on compensating evidence with the CPA firm early (scanner output, internal reviews, remediation tracking), and do not assume it is interchangeable with exploit testing.
- Package artifacts in the order the auditor reads them: report, ticket trail, retest proof, control mapping.
- Keep the evidence close to the team that fixes the issues so the next cycle is faster.
What I recommend
Start with scope, not tooling. Pick a test whose output engineers can act on and auditors can read quickly: findings with proof, tickets, a retest, all dated inside the observation window. If you can show that chain for every critical finding, the pentest question in your SOC 2 audit is settled.
Fleuret runs agentic pentests on web apps, REST and GraphQL APIs and external infrastructure, with results in hours, a replayable proof of concept per finding, findings pushed to Jira, Linear or Slack, a retest after the fix and a signed report on demand.
Sources
- SOC 2 penetration testing requirements (2026), Stingrai
- SOC 2 reports and penetration tests, Cloud Security Alliance, 2024-02-02
- State of Attacks 2025, Fluid Attacks, 2025
- SP 800-115, Technical Guide to Information Security Testing and Assessment, NIST
Your turn.
En garde.
The Free plan runs SAST, DAST, SCA and secret scanning for 2 users and 10 repos, no call needed. Several apps or an audit deadline? Book a demo instead.
