# SOC 2 and AI coding agents: reports, SSO, SCIM and audit logs compared

Which AI coding agents have a SOC 2 Type I or Type II report, where SSO, SCIM and audit logs come in, and what your own SOC 2 audit asks about agents.

- Canonical: https://replicas.dev/resources/ai-coding-agents-soc-2

A security review of a coding agent usually asks three things: does the vendor have a SOC 2 report, can we manage access through our identity provider, and can we see what the agent did. This guide answers those for eight coding agents, and lists what your own auditor will want to see once agents are writing your code.

Vendor terms change often, so every claim about a vendor below links to its own page. We checked them all on November 4, 2026.

## SOC 2 Type I and Type II

A SOC 2 report is an independent auditor's opinion on a company's controls for security and, optionally, availability, processing integrity, confidentiality and privacy ([AICPA](https://www.aicpa-cima.com/topic/audit-assurance/audit-and-assurance-greater-than-soc-2)). There are two types:

| Report | What the auditor checks | What it tells you |
|---|---|---|
| Type I | That the controls are designed well, on one date | The controls exist |
| Type II | That the controls worked, over a period of several months | The controls were followed |

Ask for the report itself, under NDA, and check which products are in its scope. A company's report doesn't cover a product that isn't named in it.

## SOC 2 reports, SSO, SCIM and audit logs by coding agent

In alphabetical order:

| Agent | SOC 2 | SSO and SCIM | Audit log |
|---|---|---|---|
| Augment Code | Type II ([Augment](https://www.augmentcode.com/security)) | SSO, OIDC and SCIM on Enterprise ([Augment](https://www.augmentcode.com/pricing)) | Audit trails ([Augment](https://www.augmentcode.com/pricing)) |
| Claude Code | Anthropic lists SOC 2 and ISO 27001 ([Anthropic](https://claude.com/solutions/enterprise)) | SAML SSO, domain capture and SCIM ([Anthropic](https://claude.com/solutions/enterprise)) | Audit logs, and a Compliance API for activity logs and chats ([Anthropic](https://claude.com/solutions/enterprise)) |
| Cursor | Type II ([Cursor](https://cursor.com/security)) | SAML or OIDC SSO on Teams and Enterprise, SCIM on Enterprise ([Cursor](https://cursor.com/pricing)) | Audit logs on Enterprise ([Cursor](https://cursor.com/pricing)) |
| Devin | Type II ([Cognition](https://devin.ai/enterprise)) | SAML or OIDC SSO and SCIM ([Cognition](https://devin.ai/enterprise)) | Every session, setting and action, with exports and an API ([Cognition](https://devin.ai/enterprise)) |
| Factory | Type I, plus ISO 42001 ([Factory](https://factory.com/security)) | Not listed on Factory's security page | Configurable audit logging ([Factory](https://factory.com/security)) |
| GitHub Copilot | Copilot Business and Enterprise are in GitHub's Type II reports ([GitHub](https://github.blog/changelog/2024-12-06-the-latest-github-and-github-copilot-soc-reports-are-now-available/)) | Through your GitHub organization or enterprise account | Agent activity in the enterprise audit log for 180 days ([GitHub](https://docs.github.com/en/copilot/reference/enterprise-administrators/agentic-audit-log-events)) |
| OpenAI Codex | OpenAI's Type II report covers the API Platform, ChatGPT Enterprise, Edu and Team, and doesn't name Codex ([OpenAI](https://trust.openai.com)) | Through the ChatGPT Enterprise workspace ([OpenAI](https://learn.chatgpt.com/docs/enterprise/admin-setup)) | Through the Compliance API ([OpenAI](https://learn.chatgpt.com/docs/enterprise/admin-setup)) |
| Replicas | Type I, with Type II in progress ([Trust Center](https://trust.replicas.dev)) | SCIM on Enterprise ([Replicas](/enterprise)) | Every configuration change, kept indefinitely, on all plans ([Replicas](/security)) |

A vendor's SOC 2 report covers the vendor's systems, not the model provider it calls. If the agent uses your own Anthropic, OpenAI or AWS Bedrock account, that provider's report covers the model.

## What your own SOC 2 audit will ask about coding agents

Once agents write code that ships, they are part of your own SOC 2 scope. Auditors test whether you follow your own controls, so an agent that bypasses them is a finding. These are the controls agents touch most:

| Control | What the auditor asks | What to set up |
|---|---|---|
| Change management | Was every change to production reviewed and approved? | Require pull request review for agent commits, and block agents from merging or pushing to the default branch |
| Attribution | Who made this change? | Commits and pull requests that name the person who started the agent, not a shared bot account |
| Access control | Who can use the agent, and which repositories and secrets can it reach? | SSO and SCIM so leavers lose access, and per-environment secrets instead of personal tokens |
| Logging | Can you show what happened and when? | An audit log of agent configuration changes, and session transcripts kept for your retention period |
| Vendor management | Have you reviewed this vendor? | The vendor's SOC 2 report, DPA and subprocessor list on file |

## Questions to ask a coding agent vendor

Ask these of any vendor, us included:

1. Can we read your SOC 2 report, is it Type I or Type II, and which products does it cover?
2. Does the agent's hosted environment, where it runs code, fall inside that scope?
3. Which plan includes SSO, SCIM and the audit log?
4. Does the audit log record what the agent did, or only what admins changed?
5. Can agents merge their own pull requests, and can we turn that off?
6. Whose model account does the agent use, and can a personal login route around it?

## Where Replicas fits

Replicas is SOC 2 Type I, with Type II in progress. You can request the report on our [Trust Center](https://trust.replicas.dev). Every workspace runs in its own VM, and security policies can attribute commits to the person who started the agent and block agents from merging. Our [security page](/security) lists every control and the plan it comes with, and our [Data Processing Addendum](/dpa) covers personal data. For HIPAA, see [AI coding agents and HIPAA](/resources/ai-coding-agents-hipaa).

If you're working through a security review, [see what Enterprise includes](/enterprise).


## FAQ

### Which AI coding agents have a SOC 2 Type II report?
As of November 4, 2026, Augment Code, Cursor and Devin publish a SOC 2 Type II report, and GitHub includes Copilot Business and Enterprise in its Type II reports. OpenAI's Type II report covers the API Platform and ChatGPT Enterprise but doesn't name Codex. Factory and Replicas have a Type I report. Ask each vendor for the report and check which products it covers.

### What is the difference between SOC 2 Type I and Type II?
A Type I report checks that a company's controls are designed well on one date. A Type II report checks that the controls worked over a period of several months. Type II tells you the controls were followed, not just that they exist.

### Does a coding agent vendor's SOC 2 report cover the model provider?
No. It covers the vendor's own systems. If the agent uses your own Anthropic, OpenAI or AWS Bedrock account, that provider's report covers the model.

### How do coding agents affect our own SOC 2 audit?
Code written by agents still goes through your change management controls. Auditors will ask whether agent changes were reviewed before reaching production, who started each change, who can use the agent and what it can reach, and whether you can show what happened. Require pull request review for agent commits, attribute commits to the person who started the agent, and keep an audit log.