Skip to main content
COOKIE NOTICECOOKIE CHOICE

We use cookies for analytics (Google Analytics) and ad measurement (Google Ads, including remarketing, and ChatGPT Ads by OpenAI). Strictly necessary cookies keep the site working. See our Cookie Policy for what each one does and how to opt out.

We’d like to use cookies for analytics (Google Analytics) and ad measurement (Google Ads, including remarketing, and ChatGPT Ads by OpenAI), and none of them is set unless you accept. Cookie Policy

Learn · SOC Reports

Carve-Out vs Inclusive Method in SOC 2

A SOC 2 description has exactly two permitted ways to handle a subservice organization. Carve-out excludes the vendor's controls from the description and from the auditor's testing, replacing them with a disclosure and a list of assumed controls. Inclusive brings them inside both — which requires the vendor's own written assertion.

The line people get wrong: carving a vendor out removes its controls from the auditor’s testing. It does not remove the disclosure, the assumed controls or your own oversight — the DC 200 definition of the carve-out method expressly includes the controls you use to monitor that vendor.

2methods permitted by DC 200
DC7the criterion governing both
300+SOC 2 engagements supported by TCSA

Plain-English explainer · DC section 200 · TSP section 100 (2017 TSC, revised points of focus 2022) · AT-C section 205 · Last reviewed August 2026

Under the carve-out method the subservice organization’s components and controls are excluded from your description and from the scope of the examination — but you must still disclose the nature of the service it performs, each applicable criterion intended to be met by controls there, and the types of controls you assumed it operates (the CSOCs). Under the inclusive method those components and controls sit inside the description and the auditor’s testing, and the vendor’s management must sign a written assertion and provide written representations. Both live in one place: criterion DC7 of DC section 200, the AICPA’s description criteria for a SOC 2 system description. Management chooses, separately for each subservice organization, and the choice turns on one question — will the vendor sign its own assertion? The examination that follows is performed under AT-C section 205, which is why the output is an attestation report carrying an opinion, and never a certificate.

Key Takeaways

  • DC section 200, criterion DC7, permits exactly two treatments for a subservice organization. There is no third option and no halfway treatment of a single vendor.
  • Carve-out excludes the vendor’s components and controls from the description and the examination, but still requires three disclosures: the nature of the service, each applicable criterion intended to be met there, and the assumed controls (CSOCs).
  • Inclusive puts the vendor’s relevant components into the description and its controls into the auditor’s testing — and needs the vendor’s written assertion, its representations, and a CPA firm independent of it as well as of you.
  • Carve-out is not an exemption: the DC 200 definition of the method includes the controls you use to monitor the vendor, and those are tested.
  • A criterion stays relevant even when the components behind it are outsourced — DC 200 uses CC6.5, secure decommissioning of storage media, as its example.
  • A heavily carved-out description is not automatically weak. Read the criteria-to-CSOC mapping, then obtain the vendor’s own report and test the assumptions against it.
What the criterion says

DC7, clause by clause

DC7 bites whenever a vendor’s controls are necessary, with yours, to provide reasonable assurance that your service commitments and system requirements are achieved. Clause (a) sets out what an inclusive description must contain, and clause (b) a carve-out one. Both govern description content only. DC section 200 says what the description must contain. The examination itself is performed under AT-C section 205, Assertion-Based Examination Engagements, which is why the output is an attestation report and an opinion — not a certificate.

DimensionCarve-outInclusive
Nature of the serviceRequired.Required.
Identity of the vendorNot a criterion. DC 200 says naming it “may be useful” to user entities. Practice is to name it.Unavoidable — its assertion is bound into the report.
Criteria involvedRequired: each applicable criterion intended to be met by controls there.Handled through the normal criteria-to-controls mapping.
The vendor’s controlsExcluded from both. Replaced by the types of controls assumed of it — the CSOCs.Described: the specific controls necessary, in combination with yours.
The vendor’s componentsExcluded. The boundary stops at the service organization.Included — relevant infrastructure, software, people, procedures and data, with the portions attributable to it separately identified.
Who is testedYou only, including your controls for monitoring the vendor.Both. The vendor’s controls fall inside the auditor’s procedures.

Two requirements that sit outside DC7 — they come from the attestation standards (AT-C sections 105 and 205) and the AICPA SOC 2 guide, not the description criteria. The first is whose management asserts: under carve-out yours is the only assertion and the only representation letter, while under the inclusive method the subservice organization’s management signs its own assertion and gives its own written representations to the CPA firm. The second is independence: the firm must be independent of you under either method, and independent of the subservice organization as well when that entity is included. Neither obligation appears anywhere in DC7, which governs what the description says rather than how the engagement is performed — a distinction worth keeping straight if you are checking this page against the standard.

Two rows in the table deserve a pause. DC7(b) requires each applicable criterion intended to be met at the vendor to be identified — not a vague nod to “infrastructure security”. And naming the vendor is guidance, not criterion: DC 200 says disclosing its identity may be useful to user entities. Practice has settled on naming it anyway, because a reader who cannot identify the provider cannot obtain the report that closes the carve-out.

Choosing

The decision, factor by factor

Management chooses per vendor, during scoping.

FactorPoints to carve-outPoints to inclusive
Will the vendor sign an assertion?No — or not without a negotiation longer than the audit. This row ends most discussions.Yes, in writing, before fieldwork is booked.
Does it publish its own SOC 2?Yes. The reader closes the loop independently — the premise of carve-out.No report exists and the dependency is too material to leave unexamined.
OwnershipArm’s-length vendor on standard commercial terms.Affiliate, captive entity, or a data centre under common ownership.
Independence exposureNeutral.Blocked if your CPA firm serves that vendor in a prohibited capacity.
Share of the delivery pathA defined layer — hosting, payments, mail delivery, alert monitoring.So much of the service that a carved-out description reads as a shell.
Period alignmentThe vendor’s period covers yours, or a gap letter is available.It reports on a cycle you cannot align to, leaving months uncovered.
Cost and scheduleNo vendor cooperation, no extra legal work, no schedule dependency.A second set of walkthroughs, a second representation letter, contract terms permitting testing.

Why the hyperscalers are always carved out. Including AWS, Azure or Google Cloud would require that provider to assert about controls inside your report, give representations to your CPA firm and open its facilities to that firm — for one customer among millions. None does. They publish their own examinations instead. Nothing in DC 200 forbids it, but the economics do. Inclusive earns its place elsewhere — a group company operating the data centre, a captive delivery entity — where cooperation is a management decision.

The part nobody escapes

What carve-out does not let you avoid

Carve-out narrows the examination, not the obligation. Five things survive.

  1. 01

    The criteria stay relevant

    DC 200 is explicit that a criterion relevant to the services provided stays relevant even if every component behind it is outsourced. Its example is an infrastructure provider responsible for wiping storage devices: CC6.5 — discontinuing protections over physical assets only once the ability to recover data has been diminished — stays relevant though that provider is carved out.

  2. 02

    Vendor management is in scope

    CC9.2 — the entity assesses and manages risks associated with vendors and business partners — sits inside the common criteria, the floor of every SOC 2. The register, the tiering, the pre-engagement assessment and the issue-handling route are yours, whichever method you chose.

  3. 03

    Monitoring is part of the method

    The DC 200 definition of carve-out names three things the description identifies: the nature of the services, the types of controls expected at the vendor, and the controls used to monitor their effectiveness. Monitoring is constituent, not optional, and maps onto CC4.1 as well as CC9.2.

  4. 04

    Their CUECs become your controls

    Nearly every vendor report contains complementary user entity controls — things it assumes you do. Multi-factor enforcement on the cloud console, key rotation, network configuration, log retention. Those are yours: they belong in your control set and must survive testing. Read the vendor’s CUEC list the same day you read its opinion, because anything on it that you are not already doing is a gap in your own control environment.

  5. 05

    The risk assessment must reach the vendor

    CC3.2 requires risks to the achievement of objectives to be identified and analysed, and DC 200 suggests describing the process for addressing risks introduced by subservice organizations and other third parties with access to customer and employee data. A register stopping at your own perimeter reads badly beside a description that carves it out.

Worked example

Four entities, one period, dated

An illustrative analytics platform, examined 1 April 2026 – 31 March 2027, Security and Availability in scope. Four entities sit in the delivery path: a cloud provider, a managed detection provider, a transactional email provider and a group-owned delivery centre. The first three are carved out, and the group entity is treated inclusively — a split DC 200 expressly permits. None of the three carved-out reports lines up with the examination period, which is the normal case rather than the awkward one.

EntityMethodIts report & periodGap vs our periodInstrument closing the gapOur control & evidence date
Cloud provider (production hosting)MethodCarve-outIts report & periodType 2, 1 Oct 2025 – 30 Sep 2026Gap vs our period6 months uncovered: 1 Oct 2026 – 31 Mar 2027Instrument closing the gapBridge letter from the provider dated 12 Apr 2027, covering 1 Oct 2026 – 31 Mar 2027. Management’s statement, not audit assurance.Our control & evidence dateAnnual inspection of the provider’s Type 2 report. Review memo dated 14 Nov 2026, 45 days after the report was published.
Managed detection providerMethodCarve-outIts report & periodType 2, calendar year 1 Jan – 31 Dec 2026Gap vs our period3 months uncovered: 1 Jan 2027 – 31 Mar 2027Instrument closing the gapGap letter from the provider dated 20 Feb 2027 for 1 Jan – 31 Mar 2027, plus the monthly service reviews already held in that window.Our control & evidence dateAnnual report review, memo dated 9 Feb 2027. Monthly alert-volume reconciliation, 12 instances in the period.
Transactional email providerMethodCarve-outIts report & periodType 1 only, as of 30 Jun 2026Gap vs our periodNo operating-effectiveness coverage at all. A Type 1 opines on design at a point in time.Instrument closing the gapCompensating security assessment completed 25 Aug 2026 and a risk acceptance signed by the CTO on 8 Sep 2026, with a 12-month re-assessment date.Our control & evidence dateAnnual report review, memo dated 6 Mar 2027 — the one deviation in this example (see below).
Group-owned delivery centreMethodInclusiveIts report & periodNo separate report. Its controls sit inside ours.Gap vs our periodNone. Its controls are described and tested across the same 1 Apr 2026 – 31 Mar 2027 period.Instrument closing the gapIts management’s own assertion, bound into Section 2, and its own representation letter to the CPA firm dated 30 Apr 2027.Our control & evidence dateNot a monitoring control — its controls appear in the Section 4 matrix with an entity column and were tested directly.
Illustrative Section 3 disclosure — carve-out

“The Company uses a third-party cloud provider to host production, a managed detection provider to monitor security telemetry, and a transactional email provider to deliver customer notifications. The description includes only the Company’s controls and excludes those of these subservice organizations. It identifies the applicable trust services criteria intended to be met by controls at those organizations and the types of controls assumed in the design of the Company’s system, presented below as complementary subservice organization controls. The Company’s commitments can be achieved only if those controls operate effectively throughout the period.”

Illustrative wording, not a quotation from an issued report.

CriteriaCSOC assumed at the vendorThe Company’s own control (tested)
CC6.4, CC6.5Restricts physical access to facilities and renders data on decommissioned storage unrecoverable.Annual inspection of the provider’s Type 2 report on physical security, in a dated memo.
A1.2Operates redundant power, cooling and network infrastructure across availability zones.Quarterly restoration into a second region, evidenced by the restore log and sign-off.
CC6.1, CC6.3Restricts its own personnel’s access to the virtualisation layer and customer control plane.Monthly review of privileged roles in our cloud accounts. Break-glass credentials in a monitored vault.
CC7.2The detection provider monitors telemetry and notifies us of qualifying alerts within contracted times.Weekly reconciliation of alert volumes against the provider portal. Monthly service review.

These CSOCs cover only the provider-side layer. The same criteria also carry tenancy-side controls — IAM configuration, network rules, logging — which are yours, sit inside the description, and are tested. That is why the same criterion number can appear both in the carve-out disclosure and in the Section 4 matrix without contradiction, and why the shared-responsibility split is the companion read to this page.

The arithmetic the CPA firm runs. The subservice-organization population for this example is four. Three are carved out, so the annual vendor-review control has a population of three. Add the pre-engagement assessment of the inclusive entity and the vendor-review register carries four lines. At that size nothing is sampled — all four are examined. Three pass without comment. One deviation is recorded: the email provider’s review memo is dated 6 March 2027, three weeks before period end and 172 days after that provider’s assurance package became available, against a control described as operating “annually, within 60 days of the vendor’s assurance package becoming available”. The control operated, but not as described — which is the definition of a deviation.

How it is dispositioned. The deviation is reported in the tests-of-controls section with its nature and extent: one of four instances, 25% of the population. Management’s explanation and remediation — a calendared review date per vendor, moved from the compliance owner’s inbox into a ticketed workflow — appear in the unaudited other-information section, conventionally Section 5. Whether the opinion is modified is the CPA firm’s judgement, and it turns on whether the underlying CSOC assumption still holds: here it does, because the security assessment of 25 August 2026 and the risk acceptance of 8 September 2026 independently supported the assumption during the window the memo was late. The group delivery centre is tested in the Section 4 matrix alongside the Company’s controls, with its own assertion bound into Section 2.

Evidence

What the CPA firm actually requests

A short list generating a disproportionate share of review comments.

RequestAcceptedComes back
The complete populationThe vendor register exported from the source system with the export timestamp and the operator’s name visible, reconciled to the cloud-account inventory, the SSO application list and payables above a stated threshold.A hand-typed list with no source system. A population that cannot be tied out cannot be sampled.
Classification rationaleA line per vendor answering the DC 200 test: are its controls necessary, with ours, to achieve our commitments?Tiering by spend or criticality. Neither is the criterion, and both let subservice organizations slip out.
CSOC derivation paperEach criterion mapped to the assumed control, traced to a named control in the vendor’s own report, with the vendor’s control reference recorded.Template CSOCs naming controls the vendor’s report does not contain, or “maintains appropriate security controls”.
The vendor’s SOC 2 Type 2The full report plus the download record proving when it was obtained — for AWS, the AWS Artifact download record showing report title and period. Read for period, opinion, exceptions, its CUECs and its own carve-outs.A SOC 3, a trust-centre screenshot or a badge. None carries the tests performed and the results — a SOC 3 is a general-use summary with no test matrix and no exceptions.
The review memoDated, named reviewer, period reviewed, opinion, exceptions considered, CUECs accepted into our control set.“Reviewed — no issues.” Undated memos. Memos dated outside the period for a control meant to operate inside it.
Exception follow-upA ticket per exception touching a CSOC we rely on, with the ticket ID quoted in the review memo, the compensating control and the risk decision.Silence. An unread exception is a finding of its own.

Population and sample by control frequency

Market practice — the AICPA prescribes no sample sizes.

Control frequencyPopulation over 12 monthsWhat the firm typically examinesThe artifact requested
Annual — vendor report reviewOne instance per carved-out entity. Four entities, population of four.All of them. At this size selection is not worth the argument.A dated review memo per entity, naming the reviewer, the report period and the exceptions considered.
Quarterly — service review against agreed levelsFour instances over a 12-month period.All four.Meeting minutes carrying the date, the attendees from both sides and the metrics discussed.
Monthly — privileged access reviewTwelve instances over a 12-month period.Commonly two to four selections.The signed review export showing the reviewer’s name, the run timestamp and the accounts removed.
Event-driven — new vendor onboardingEvery onboarding in the period. Often single digits.All of them under roughly ten, a selection above that.The completed security assessment plus the approval ticket, with the approver and date visible.

Population and sample, in plain terms

The attestation standards prescribe no sample sizes. Firms set their own approach, most carrying over attribute-sampling conventions from audit practice — commonly a 90% confidence level on larger populations. This is practice, not a requirement. The population is the number of times the control should have operated: an annual review across six carved-out entities is a population of six, and at that size firms usually examine all of them. Sampling is rarely the problem. The population is. A register that cannot be tied to an independent source yields samples nobody can rely on, and the first review comment of the engagement is almost always about how the list was built rather than what was selected from it.

Testing

What the auditor tests about your monitoring

Under carve-out the auditor tests you, not the vendor. Seven procedures, and the deviation each one most often produces.

ProcedureWhat is inspectedPopulationThe exception that usually results
Inspection of the inventory and classification rationaleThe register export and the one-line rationale per vendor, tied to the cloud-account inventory, the SSO application list and payables above a threshold.One register, walked at scoping and re-checked at fieldwork.A vendor sitting in the payables listing that never reached the register — the description is incomplete, which is a fair-presentation problem.
Inspection of each vendor report obtained in the periodReport period, opinion, every exception, the vendor’s own CUECs and its own carve-outs.One per carved-out entity.A report downloaded and never read — no annotation, no CUEC extraction, nothing carried into the control set.
Inspection of the dated review memo, plus enquiry of the named reviewerDate, reviewer, period reviewed, opinion reached, exceptions considered, CUECs accepted.One per entity per period.A memo dated inside fieldwork rather than inside the period, so the control did not operate throughout.
Re-performance of the CSOC mappingEach assumed control traced to a named control in the vendor’s report, and to the criterion the description claims it meets.Every criterion the description lists as intended to be met at a subservice organization.A CSOC describing a control the vendor’s report does not contain — the assumption has no support.
Inspection of exception follow-upThe ticket for each exception touching a CSOC, the compensating control, the risk decision and who signed it.Every exception in every vendor report obtained in the period.An exception read and left undispositioned. Silence is itself the finding.
Inspection of contractsThe security schedule, the incident-notification window and the change-notice terms the monitoring control depends on.One contract per carved-out entity.A control that says the vendor notifies you of incidents within a window the contract never required.
Inspection of the risk assessmentEvidence that risks introduced by third parties were identified and analysed during the period, not at renewal.One assessment cycle falling inside the period.A register that stops at your own perimeter while the description carves four entities out.

Timing produces more deviations here than substance. An annual review run three weeks before fieldwork has not operated throughout a twelve-month period, and a memo dated after period end cannot evidence a control the description places inside it. The cheap fix is calendaring each vendor review against that vendor’s own publication cycle rather than against your audit calendar — and writing the frequency into the control description in terms you can actually meet.

For the reader

Reading a heavily carved-out description

Modern software is assembled, so a long carve-out list is not itself a weakness. What separates a strong carved-out description from a thin one is whether the exclusions are mapped or announced. Five checks, each of which you can perform on the PDF in front of you.

  • Search the report for the exact heading “Complementary Subservice Organization Controls”. By convention the list sits inside Section 3, management’s description of the system, usually as a table immediately after the subservice-organization disclosure. If the phrase returns no hit, the disclosure is either missing or buried in prose.
  • Check that the description states that the service organization’s commitments and system requirements can be achieved only if those complementary controls are suitably designed and operating effectively along with the entity’s own controls. That statement is what makes the assumption explicit rather than implied.
  • Read the criteria column of that table. DC7(b) requires each applicable criterion intended to be met at the vendor to be identified — a table listing vendors with no criteria attached has not met the criterion.
  • Look for the specific mismatch: a criterion listed as met by controls at the vendor that also appears in the Section 4 matrix as tested at the service organization, with no explanation of the split. One of the two placements is wrong, and the question is worth asking.
  • Compare the carve-out list against the sub-processor list in the data processing agreement. They legitimately differ in both directions — but a hosting provider on one and absent from the other is worth a question, and so is what is left inside the boundary: if access, change, incident response and encryption are tested at the service organization, the report is substantive whatever the carve-out count.
“A carved-out opinion is conditional on controls nobody tested. Closing that loop is the reader’s job.”
Objection handling

What a buyer’s security team pushes back on

Each is reasonable, and each has a factual answer.

The objectionWhat is actually trueHow to answer it
“You carve out your cloud provider, so the report covers nothing.”Your access, change, incident, encryption and oversight controls are all tested. Only the provider’s control over its own facilities is excluded.Walk the criteria mapping: met by you alone, by a combination, or at the provider. DC7 requires that list.
“Why not use the inclusive method?”It needs the vendor’s assertion, representations and consent to be tested. Hyperscalers give none of the three.Say so plainly, and note the method is chosen per entity — you may carve out the cloud and include an affiliate.
“Send us your cloud provider’s report.”SOC 2 reports are restricted-use. You are rarely licensed to redistribute a vendor’s report.Point them to the provider’s portal with the exact report name and period.
“Your CSOC list is boilerplate.”Often true. DC 200 requires CSOCs to be complete, accurately described and relevant to the achievement of your service commitments and system requirements.Rewrite it from the vendor’s report and delete CSOCs unrelated to your delivery path — DC 200 adds that only CSOCs specific to the services provided should be described.
“An annual report review is not monitoring.”DC 200 lists reconciling output reports, service reviews against agreed levels, site visits and monitoring of external communications alongside report inspection.Describe the whole set, including incident-notification handling and the re-assessment a vendor security event triggers.
“Your sub-processor list and your carve-out list do not match.”They are built on different tests, so they legitimately differ in both directions — a sub-processor may not be a subservice organization, and vice versa.Show both tests side by side and name the entities that appear on one list only, with the reason.
“What happens if your cloud provider is breached during the period?”A known incident at a subservice organization is a disclosure question under DC 200’s incident guidance, regardless of the method elected. Carving a vendor out does not carve out an incident there.Point to the disclosure in the description and to the evidence your own incident-handling and re-assessment controls produced. It is not a scoping question.

Suggested wording — paste-ready questionnaire replies

When they say the carve-out empties the report

“Our examination carves out the hosting provider, which is the standard treatment permitted by criterion DC7 of DC section 200. The description identifies every trust services criterion intended to be met by controls at that provider and the specific controls we assume it operates. All access, change management, encryption, incident response and vendor-monitoring controls remain inside the described system and were tested by our CPA firm across the full period.”

When they ask for the provider’s own report

“SOC 2 reports are restricted-use documents and we are not licensed to redistribute our provider’s. You can obtain it directly and at no cost from the provider’s trust portal — the report you want is its SOC 2 Type 2 for the period covering ours. We are happy to confirm in writing the exact report title and period we reviewed, and the date of our review.”

When the sub-processor list and the carve-out list differ

“The two lists answer different tests. A sub-processor is an entity processing personal data on our instructions, which is a data-protection classification. A subservice organization is an entity whose controls are necessary, in combination with ours, to achieve the commitments made about this system. Entities can sit on one list and not the other in both directions, and ours do. We can walk you through each difference and the reason for it.”

Suggested wording only. Check each against your own description and report before sending.

In SOC 2 readiness work, Tranquility Cybersecurity builds the subservice-organization inventory, the method decision per entity, the criteria-to-CSOC mapping and the monitoring evidence before the period opens, and coordinates the engagement with the licensed CPA firm that performs the examination and issues the opinion. Engagements are quoted after scoping, at a fixed fee from $2,500 for a Type 1 and $3,000–$10,000 for a Type 2, all-in, including the independent CPA attestation.

Carve-Out vs Inclusive — Common Questions

What each method requires, who decides, and what changes for the reader.

What is the carve-out method in a SOC 2 report?
It excludes the components of a subservice organization’s system used to serve you from both the description and the scope of the examination. Three disclosures survive: the nature of the services performed, each applicable criterion intended to be met by controls there, and the types of controls assumed of it — the CSOCs. A fourth thing survives that people miss: the DC 200 definition of the method itself includes the controls you operate to monitor the vendor, so those sit inside the described system and the CPA firm tests them like any other control. In practice that means a register, a dated review of each vendor’s report, and documented follow-up on every exception that touches a CSOC you rely on.
What is the inclusive method, and why is it rare?
The relevant aspects of the vendor’s infrastructure, software, people, procedures and data become part of your described system, with the portions attributable to it separately identified, and its controls fall inside the auditor’s procedures. Mechanically the report changes in six places: two assertions rather than one, two representation letters, a components section covering both estates, an entity column in the control matrix, no CSOC disclosure for that entity, and an independence requirement that must clear the vendor as well as you. It is rare because every one of those needs the vendor’s cooperation as a matter of governance, not commerce — which is why it appears almost exclusively between entities under common ownership.
Who decides which method to use?
Service organization management decides, per subservice organization, during scoping. DC section 200 states that an organization using several subservice organizations may use carve-out for some and inclusive for others, so a mixed treatment inside one report is expressly permitted rather than a workaround. The CPA firm does not choose the method, but it will challenge a classification it disagrees with. If a vendor you treated as an ordinary supplier is in fact a subservice organization, the description omits a required disclosure — a fair-presentation problem resolved by amending the description before issue, or, failing that, by a modified opinion. Settle classification before the period opens, not during fieldwork.
How do I tell a strong carve-out from a weak one?
Open the report and run five checks. Search for the heading “Complementary Subservice Organization Controls” — if it returns nothing, the disclosure is missing or buried. Confirm the description states that commitments can be achieved only if those controls operate effectively alongside yours. Read the criteria column: DC7(b) requires each applicable criterion intended to be met at the vendor to be named, so a vendor list with no criteria attached has not met the criterion. Look for a criterion claimed at the vendor that also appears in the control matrix as tested in-house with no explanation. Then check what is left inside the boundary — access, change, incident response and encryption tested in-house make a report substantive whatever the carve-out count.
What are CSOCs and where do they appear?
Complementary subservice organization controls are the controls management assumed, in designing the system, would be implemented by the subservice organization and which are necessary — with your own — to achieve your service commitments and system requirements. DC 200 requires them to be complete, accurately described and relevant to the achievement of those commitments and requirements, and adds that only CSOCs specific to the services provided should be described. By convention they appear as a table inside Section 3, mapped to criteria, with the description stating that commitments can be achieved only if they are suitably designed and operating effectively along with your controls. They are the mirror image of the CUECs your own report places on your customers.
Do I still need vendor management if I carve everything out?
Yes — more of it, not less. CC9.2 places vendor and business-partner risk inside the common criteria, so it applies whichever method you elected, and the DC 200 definition of carve-out itself includes the controls used to monitor the vendor. In practice: a maintained register tied to an independent source such as the cloud-account inventory and payables above a threshold, risk-based tiering, a pre-engagement assessment, a dated annual review of each vendor’s report, and a documented route for handling issues. DC 200’s implementation guidance also names reconciling output reports, evaluating performance against service-level agreements, site visits and monitoring external communications as qualifying monitoring activities — describe the whole set, not just the report review.
What if a subservice organization has no SOC report?
Carve-out is still permitted. Nothing in the description criteria conditions it on the vendor holding a report. What changes is the evidence behind your monitoring control, which now rests on a completed security assessment, an independent penetration-test summary, contractual security schedules and a documented risk acceptance signed by someone with authority to accept it. The CPA firm tests that package closely, because it alone supports the CSOC assumption — expect requests for the assessment questionnaire and the responses, the date it was completed, the risk decision and its re-assessment date. The same applies where the vendor holds only a Type 1: design at a point in time cannot evidence operating effectiveness across your period.
What happens if the vendor’s report is qualified, or an exception hits a CSOC?
Work in three steps. First, determine whether the exception or the basis for the qualification actually touches a CSOC in your mapping — plenty of exceptions sit in areas you place no reliance on, and the answer is a documented conclusion rather than escalation. Second, if it does touch one, record the compensating control and the risk decision inside the period, signed by someone with authority. Third, expect your CPA firm to test that follow-up as part of your monitoring control. The important point: this is a finding about your monitoring, not a change to your scope. The vendor stays carved out and the method does not change because its report came back imperfect.
Can I switch from carve-out to inclusive between periods?
Yes. The method is elected per entity per period, so a vendor carved out last year can be included this year and the other entities in the same report are unaffected. Two consequences follow. First, the change alters the described system: the vendor’s relevant components come inside the boundary, its controls enter the matrix, and its CSOC disclosure disappears because nothing is left as an assumption. Second, for a type 2 the description discloses relevant details of significant changes during the period, so a switch that happens mid-period is disclosed rather than silently applied. The practical constraint is unchanged — the vendor must sign an assertion, and if it will not, the switch cannot happen.
Does any of this differ for SOC 1?
The methods work the same way, but the standards behind them differ and it is worth stating precisely. A SOC 2 examination is performed under AT-C section 205, Assertion-Based Examination Engagements, against the trust services criteria in TSP section 100, with description content governed by DC section 200. A SOC 1 is performed under AT-C section 320, whose description requirements are written into the section itself rather than a separate DC document, and framed against control objectives relevant to user entities’ internal control over financial reporting. Both are attestation engagements producing an opinion, never a certification. The CSOC concept, the vendor’s assertion under the inclusive method and the surviving monitoring duties are common to both.

Written By Expert Auditors

Surendra Pal SinghChief Information Security Officer & Data Protection OfficerCISODPOCISAMCSEITILISO 27001 Lead AuditorISO 27701 Lead AuditorISO 42001 Lead Auditor
Saundhi ChauhanLead AuditorISO 27001 Lead AuditorISO 27701 Lead AuditorISO 42001 Lead Auditor
Last reviewed: August 2026 • Content verified by certified lead auditors

Get in touch

Book a free consultation or send us your requirements. We respond within 24 hours.

Send us your requirement.

01Your details
02What is it about?
SOC 2
03Send
Asking about: SOC 2