Home / Solutions

Keep agent changes inside SOC 2 controls

AI governance solutions should check SOC 2 commitments in the systems that coding agents change.

Updated November 12, 2026

AI governance solutions control and record AI use

AI governance solutions include AI governance tools and services that find uses, enforce policy and keep evidence. ZeroTB maintains SOC 2 controls in the systems coding agents change.

  • Serve audited companies. We support US software, AI, fintech and healthtech companies with 25 to 250 people

  • Maintain framework commitments. Our work covers SOC 2, ISO 27001 and HIPAA after the first audit

  • Check connected systems. We read GitHub, AWS, Google Workspace, identity providers and endpoints

An approved coding agent works under the repository permissions and release settings each team connects it to.

AI governance challenges start once agents are approved

AI governance challenges continue after approval: a repository loses required review, another starts without protection, or an account gains permissions the approval never covered.

Compliance interprets the requirement, engineering controls the repository and security reviews access. The gap can pass between them.

AI governance software checks systems against AI policy

AI governance software tests the conditions an AI use policy describes, such as required repository review. The policy sets permitted work and exception owners.

The NIST AI RMF treats risk management as work throughout an AI system's life. AI governance frameworks covers that wider program.

AI governance tools differ by what they control

AI governance tools cover distinct jobs. AI governance platforms may bundle several, from finding shadow AI to restricting an agent's next action.

TaskAI inventory and shadow AI discoveryPolicy and risk workflowsRuntime guardrailsData classification and DLPZeroTB
What it controlsAI tools in use and their ownersAssessments and approvals of AI usesActions agents can take at run timeSensitive data reaching AI toolsSOC 2 controls in repositories, cloud and accounts
When it actsWhen a new AI tool appearsBefore a use is approvedOn each request or actionWhen data movesDaily checks and repair of each gap
What a reviewer readsAn inventory of uses and ownersRisk assessments and decisionsLogs of allowed and blocked actionsData policy eventsA signed Verified Control Record per repair
Fits whenAI use is unknownUses need sign-offAgents need limits on actionsSensitive data reaches AI toolsSOC 2 controls drift as agents change systems

Start with inventory and owners when AI use is unknown. Once uses are recorded, establish who repairs failed controls.

Agentic AI governance limits permissions and approves changes

Agentic AI governance starts with what connected tools permit. Reading a repository requires different authority from merging code or changing access.

OWASP excessive agency guidance calls for limits and human approval of the system, action and expected result. Owners decide exceptions.

AI agent governance needs an audit trail linking actions to approvals and resulting states. Test claims to record every change.

Check repository actions, identities, resources, local commands and connected services. New repositories can escape scope. Unknown, stale and disconnected checks remain visible.

ZeroTB repairs the controls that agents break

ZeroTB restores SOC 2 controls through daily checks, repairs, human approval and verification. Checks examine systems without connecting to coding agents.

Runbooks, pull requests, API actions or guardrails address recurring gaps. The signed Verified Control Record keeps commitments, before-and-after states, actions, approvals and verification together.

Example of restoring required review

This example follows a repository used by coding agents after required review disappears. The record shows the approved correction and the check that follows.

Required review is back in place
A review restoration record: a repository used by coding agents merging without review, the engineering owner's approved correction, a later check confirming the setting applies, and the retained observations and approval.
  1. Observed review gap. Changes can pass without the review required for production changes.

  2. Approved correction. The engineering owner approves restoring required review on the branch.

  3. Verified review setting. A later check confirms that the setting applies to the branch.

  4. Retained repair evidence. The record keeps both observations, the approval and the completed action.

Test each vendor on one SOC 2 commitment

Name the commitment

Follow a SOC 2 commitment already held, such as reviewing production changes.

Watch the check

Ask the vendor to test the repositories coding agents use.

Follow the finding

Establish who investigates, drafts the repair and checks the result.

Test the approval

Confirm that approval covers the specific action about to happen.

Open the record

Read the requirement, observations, approved action and later check together.

FAQ

What is the difference between AI governance tools and an AI governance platform?

AI governance tools cover individual jobs such as inventory, risk decisions, runtime limits or data protection. An AI governance platform bundles several of those capabilities.

What does AI governance compliance mean for a SOC 2 company?

AI governance compliance means SOC 2 commitments still hold as coding agents change repositories, cloud settings and access. AI governance frameworks explains the wider program, including NIST AI RMF.

Is ZeroTB an AI governance platform?

ZeroTB combines software and a team to maintain SOC 2, ISO 27001 and HIPAA controls after audits. Inventory and runtime guardrail tools cover other AI governance jobs.

Does ZeroTB connect to our coding agents?

ZeroTB connects to no coding agent. ZeroTB checks the systems agents change: GitHub, AWS, Google Workspace, identity providers and endpoints.

What access does ZeroTB need to our repositories and cloud accounts?

ZeroTB starts read-only. Write access is scoped by action class and revocable. Your team reviews material changes as pull requests. Direct API actions are opt-in and reversible.

Start with the Pilot

The eight-week Pilot maps up to 25 commitments and restores two real control gaps, with verified results and signed records.