Buying a compliance dashboard is often the first step a software team takes when an enterprise customer asks for SOC 2. Those platforms are useful enough for collecting screenshots, though an independent CPA firm is not testing whether you subscribed to compliance software. They are testing whether your team actually followed the operational controls described in your system description. A SOC 2 Type I report evaluates how those controls are designed as of a single point in time. A SOC 2 Type II report looks across a three- to twelve-month observation period, sampling real tickets, pull requests, offboardings, and alerts.
How do capable engineering teams still end up with exceptions listed in their final report? More often than not, someone adopted an overly aggressive policy template downloaded from the internet (promising 48-hour remediation on medium-severity findings or weekly manager reviews that nobody runs) and forgot about it until audit fieldwork started. Writing realistic controls that match how your engineers ship software, then automating the evidence underneath them, avoids most of those self-inflicted findings.
Scope Only the Trust Services Categories Your Customers Require
Under the AICPA Trust Services Criteria, Security (the Common Criteria, CC1 through CC9) is the only mandatory category in every SOC 2 report. First-time teams sometimes assume checking every optional category makes the report look stronger. Usually it just widens the sample population without helping close a single extra contract.
Check your signed customer agreements and active security questionnaires before adding optional categories. Availability (A1) belongs in scope when enterprise contracts carry uptime SLAs, RTO/RPO commitments, and annual disaster recovery restore testing. Confidentiality (C1) fits platforms that store proprietary business files, legal agreements, M&A data rooms, or customer source code requiring documented retention and disposal proof. Processing Integrity (PI1) and Privacy (P1) sit narrower still, reserved for systems where financial transaction calculations or consumer privacy notice and consent workflows sit at the center of procurement requests (see how international buyers compare these in SOC 2 vs. ISO 27001).
Where Logical Access Sampling Gets Strict (CC6)
Logical access controls receive the heaviest sampling during Type II fieldwork. Auditors ask HR for complete hire, transfer, and termination rosters across the observation period, then cross-check every departed employee and contractor against corporate SSO, AWS, Azure, or GCP accounts, GitHub, and production databases. If your policy commits to revoking access within 24 hours of termination, a single contractor left active in GitHub for five days becomes a sampled exception. Automating SCIM deprovisioning covers SSO-connected apps, while a structured quarterly user access review, with an independent executive reviewing administrator roles, catches systems outside SCIM.
How privileged users authenticate matters just as much as how promptly you offboard them. Standard 6-digit TOTP authenticator apps and push prompts satisfy basic MFA checkboxes in many policies, though neither stops an Adversary-in-the-Middle reverse proxy from capturing a live session. Require MFA across 100% of workforce logins and production bastion access, and move privileged cloud, Git, and identity-provider administrators onto hardware-backed FIDO2/WebAuthn security keys, device-bound passkeys, or Windows Hello and Platform SSO (covered in our Entra ID phishing-resistant MFA guide). In production databases and Linux servers, retire shared logins altogether so every session ties back to one named engineer via IAM database authentication, SSM Session Manager, or short-lived bastion certificates.
Peer-Reviewed CI/CD and Late-Night Hotfixes (CC8)
During change-management testing, auditors pull the complete list of production deployments across the observation period and sample 25 to 45 changes. For each sample, they check that automated tests passed and that an independent peer—never the commit author—approved the pull request prior to merge. You can verify your GitHub branch protection rules directly from the CLI:
# Verify GitHub main branch requires peer review, enforces rules on admins, and dismisses stale approvals
gh api repos/EXAMPLE-ORG/EXAMPLE-REPO/branches/main/protection \
--jq '{
enforce_admins: .enforce_admins.enabled,
required_approvals: .required_pull_request_reviews.required_approving_review_count,
dismiss_stale_reviews: .required_pull_request_reviews.dismiss_stale_reviews
}'
Pay close attention to enforce_admins. Leaving repository owners able to bypass pull requests for "one-line config tweaks" is probably the single fastest way to fail CC8 sampling.
What happens late at night during a SEV1 outage when a fix has to ship immediately and a second reviewer is asleep? Rather than disabling branch protection in the dark and hoping that commit never gets sampled, write an explicit emergency-change clause into your change management plan. Allow break-glass deployment tied to an active incident ticket, provided automated CI verification and retrospective peer code review close within 24 to 48 business hours. Keep production data out of staging environments too: separate production and non-production VPCs, cloud accounts, and KMS keys, and seed staging with synthetic data instead of unmasked customer database dumps.
Setting Policy Timelines Your Team Will Keep (CC3, CC7, CC9)
AICPA criteria call for documented risk management, vulnerability remediation, monitoring, and continuity processes, yet they rarely dictate exact numeric SLAs. Your own policies set those clocks, and once you write a timeline down, the CPA firm holds you to it:
| Control Area | Practical Engineering Policy Baseline | Primary Evidence Sampled by CPA Firm |
|---|---|---|
| Risk register (CC3.2) | Review and update at least annually and upon major architecture changes using a consistent 5x5 matrix | Dated security risk register with named owners, mitigation tickets, and time-bounded executive sign-off on accepted risks |
| Vulnerability management & pentesting (CC7.1) | Achievable remediation targets (for example, 7-14d Critical exploitable, 30d High, 90d Medium) plus annual manual pentest | Container and dependency scan exports, closed remediation tickets within SLA, and penetration test report with verified retest |
| Centralized logging & alerting (CC7.2, CC7.3) | Append-only cloud audit, auth, and database DDL logs retained for 365 days with paged alerts on critical events | CloudTrail, Entra, or SIEM retention configuration screenshots plus sampled incident tickets (incident runbook) |
| Backups, DR drills & vendor CUECs (A1, CC9.2) | Immutable encrypted backups, annual timed restore drill, and annual critical-vendor SOC 2 review | Dated DR restore drill record plus vendor review mapping Complementary User Entity Controls (CUECs) |
Check Your Own Populations Six Weeks Early
CPA auditors do not take exported spreadsheets at face value. They verify Information Provided by the Entity (IPE) by comparing your submitted deployment lists, employee rosters, and ticket exports against raw GitHub commit logs, HR payroll dates, and live cloud settings. Six weeks before your audit window closes, pull a quick internal sample (five random deploys, three offboardings, and five vulnerability tickets) to confirm every timestamp and approval stands on its own. If your team wants hands-on engineering help closing technical gaps before bringing in an auditor, take a look at our security and compliance engineering and PCI DSS consulting services.