← All posts
SOC 2 for serverless SaaSserverless SOC 2 scopeAWS Lambda SOC 2 evidence

SOC 2 for Serverless SaaS: Trace the Work Beyond the Function

A managed function does not cover your SaaS SOC 2 work. Map triggers, permissions, releases, logs, failed events, and source evidence across the service.

A serverless service still has code, permissions, data paths, releases, and operating records to review.
A serverless service still has code, permissions, data paths, releases, and operating records to review.

For SOC 2, a serverless SaaS team still needs to describe the service it runs and show how its controls work. The cloud provider operates the underlying function platform. Your team owns its application code, permissions, configuration, data paths, releases, and responses to failures. Trace a real customer request and a background job through those paths before choosing evidence. A provider’s SOC 2 report can help with provider-side controls, but it is not a report on your SaaS.

What changes for SOC 2 when the app is serverless?

The AICPA SOC 2 Description Criteria ask management to describe the system used to provide its services, including relevant infrastructure, software, people, procedures, and data. They do not prescribe a server type or a fixed set of cloud screenshots. Management and the independent CPA firm agree on the examination scope, and the CPA firm evaluates the controls and evidence.

AWS’s Lambda security overview places the underlying infrastructure and operating system with AWS, while the customer remains responsible for its function code, libraries, resource configuration, and identity and access management. Other managed services have their own responsibility split. Check the actual services and provider terms in your architecture. The cloud provider report guide explains how to review the provider’s report without presenting it as yours.

Draw the event path before the control list

Start with one customer request, such as uploading a file. Then follow one asynchronous step, such as a queue message that starts a processing function. For each step, record the actor, trigger, function or service, data touched, permission checked, output, owner, and source system that could prove what happened. This is a working scope map, not a claim that every path passes a control.

Path to map Question for your team Source to inspect
Entry and identity Which caller may trigger the request, and how is its customer context checked? API authorization, identity settings, test result
Function and data Which code version runs, what data can it read or write, and which role grants that access? Deployed function version, execution role, data permissions
Event source Who can publish the event, and can the consumer trust its target and customer context? Event source settings, producer review, authorization test
Failure path What happens after a timeout, retry, duplicate event, or exhausted attempt? Retry settings, queue or destination, alert and incident record
Release path Which reviewed change and build reached the live function and its configuration? Source revision, build result, deployment record
Support path Who can inspect logs, replay work, or alter production settings? Access grant, approval, audit event

This map makes a serverless service less mysterious. The function may be small, but the identity, queue, data store, CI pipeline, alert, and support tool can all affect the customer-facing service. Use the SaaS scope guide to connect these components to the report boundary and buyer commitments.

Check the log you think you have

Do not use “we have CloudTrail” as proof that every customer action is logged. AWS’s Lambda CloudTrail guide separates management events, such as changes to function configuration, from invocation data events. Most data events are not logged by default, and the standard Event history shows management events for a limited window. Lambda function logs are a separate source; AWS’s logging guide explains their destinations and the execution-role permissions they need.

Choose the records that support each actual control. A release control might need the approved source change, build, deployment, and configuration history. An incident control might need a relevant alert, investigation, and response record. An access control might need role configuration, grants, and review. Check retention and access for each source before a proposed Type 2 period. The SOC 2 audit log guide helps separate what to record from how long to keep it. Avoid putting raw customer payloads, credentials, or personal data that may need erasure into Git.

Prove what happens when the event fails

An asynchronous job needs a failure story because a successful invocation does not show what happens to the rest of the work. AWS’s Lambda retry guidance describes retries and options to retain discarded events or failed-invocation records. Queue and stream event sources can have different retry behavior, so inspect the exact invocation path in your service.

Run a safe test with synthetic data. Send one valid event, then an event that fails in a controlled way. Record the trigger, function version, retry or failure destination, alert, owner response, and final disposition. If duplicate delivery could change customer data, test the behavior that prevents an unwanted second action. Treat a missing destination or alert as a gap to resolve against your commitments and risk assessment, not an automatic SOC 2 failure. The startup risk assessment guide helps connect that decision to the service you actually run.

Keep code, deployment, and operation separate

Infrastructure as code can show the configuration your team proposed and reviewed. It does not, by itself, show which version reached production or whether an alert, review, or recovery procedure operated during the period. For one change, connect the approved source revision to the built artifact, deployed function and settings, verification result, and any exception. The change management evidence guide covers that chain in detail.

For a Type 2 examination, keep dated records from the period the CPA firm examines. Do not backfill an older week with today’s configuration screenshot. The AICPA criteria distinguish a Type 1 examination of design at a date from a Type 2 examination that also addresses control operation during a period. Your CPA firm decides which evidence is sufficient for the agreed scope.

FileGRC is a Git-native GRC workspace for SOC 2 work. In the FileGRC workspace, JSON holds structured records, Markdown holds long-form work, and Git supplies the change history. Your team can link the service map, control owners, risk decisions, evidence references, and gaps so an agent can inspect the whole program and prepare sourced answers. Starter records are proposals, not compliance claims. The cloud, identity, CI, monitoring, and backup systems operate controls and produce source evidence. Your team reviews program work and customer answers; an independent CPA firm performs the examination and judges evidence sufficiency.

Open source · MIT

Run your SOC 2 program as files in Git.

Keep policies, controls, work, and evidence indexes in a repository your team and agents can inspect. Add optional FileGRC Autopilot email and Slack reminders to keep work moving.

Frequently asked questions

Does using serverless functions mean our SaaS is already covered by our cloud provider's SOC 2 report?

No. The provider's report covers the provider system and controls described in that report. Your company's code, configurations, access decisions, data flows, operations, and customer commitments remain part of the service you provide. Ask the buyer what report it needs and review your proposed scope with an independent CPA firm.

Does SOC 2 require a serverless startup to log every function invocation?

There is no universal SOC 2 rule to log every invocation. Choose logs and retention based on the service commitments, risks, control design, and evidence needed for your scope. In AWS Lambda, CloudTrail management events and optional invocation data events are different sources; function application logs are another source. Verify each setting rather than assuming one log covers all three.

What should we record for an asynchronous serverless job?

Map the producer, event source, function, data store, retry and failure paths, owner, and source records. Show how the team detects and handles an event that fails or expires. The exact controls depend on the invocation method, customer promises, and system design; discuss the proposed approach with the CPA firm.

Can infrastructure as code alone prove our serverless controls operated?

No. Infrastructure as code shows a proposed or reviewed configuration. Compare it with deployed settings and dated source records such as access grants, release results, alerts, incident work, and restore tests where those controls apply. A Type 2 examination also considers operation over its specified period.

Can FileGRC pull Lambda evidence and decide whether it passes SOC 2?

No. FileGRC is a Git-native GRC workspace. It keeps structured records in JSON, long-form work in Markdown, and change history in Git. Your team collects or references source evidence from cloud and other systems, reviews gaps, and works with the independent CPA firm, which judges evidence sufficiency and performs the examination.