AI can help run the company.It cannot quietly change it.

Who gets in. What agents can do. Who decides.

Clear answers on access, AI, and approval — before you connect anything.

1. 01

   What data can Checks & Balances AI access?

   Only what you put in or connect.

   That may include your strategy, decisions, roadmap, files, comments, and approved connected systems. Access is limited by the permissions you grant through OAuth, API keys, or MCP. We do not get blanket access to your company.

   • Strategy, decisions, roadmap, files, and threads are stored inside your company workspace
   • Connected tools are limited by the permissions you approve
2. 02

   Who can access the workspace?

   Access is invite-only. Someone must be a member of your company workspace before they can see strategy, decisions, roadmap, files, or agent work.

   Owners and Admins manage members, roles, and settings. Members can do day-to-day work. Viewers can look, not change.

   • Invite-only workspace membership
   • Roles: Owner, Admin, Member, and Viewer
   • Admins control who can join and what they can do
3. 03

   Where is company data stored?

   Checks & Balances AI uses cloud infrastructure in the United States. Company data is logically isolated by workspace and encrypted in transit and at rest.

   Pilot customers can review our current infrastructure, subprocessors, data flow, retention, and deletion practices with our team before connecting sensitive systems.

   • Google Cloud infrastructure in the United States
   • Tenant isolation: one company workspace, separate from all others
   • No cross-tenant access to strategy, decisions, files, or threads
   • Encrypted data in transit and at rest
   • Owners can schedule workspace deletion
4. 04

   Which AI models can receive company context?

   Selected AI providers process the context needed for the task — not a full dump of your company.

   Checks & Balances AI does not use customer data to train its models. Providers are used under their business API terms.

   • Task-scoped AI context instead of full-company dumps
   • Customer data not used to train Checks & Balances AI models
   • Admin control over which models agents may use
5. 05

   Who can approve company changes?

   Agents can draft, compare, check, and suggest. They cannot lock or approve company direction.

   Only a human can approve a decision. There is no auto-approve path.

   • Only humans can approve and lock decisions
   • Agents cannot silently change approved company direction
   • Decision history with approval trail
6. 06

   What can agents change in connected tools?

   You choose which systems to connect. Some integrations are read-only by design; others may support writes once connected.

   A clear per-system policy — for example CRM read-only, write, or off — with human approval before external writes is still rolling out.

   • You choose which systems to connect
   • Read-only patterns for some integrations
   • Per-system read / write / off policy (e.g. CRM)Coming soon
   • Human approval before external writesComing soon
7. 07

   Can we see what happened?

   Yes. The product is built around traceability.

   You can see what changed, who approved it, what evidence supported it, and which downstream work was affected.

   • Version history for strategy, decisions, artifacts, and roadmap work
   • Linked evidence and reasoning on important decisions
   • Change trail for approved company direction
   • Exportable audit reportsComing soon
8. 08

   What happens if something is wrong?

   Bad state inside Checks & Balances AI should be visible and fixable. We can restore approved workspace objects we own — strategy, decisions, and linked work.

   We cannot magically roll back every connected tool. External revert depends on that system’s own history, permissions, and your company policy for what agents are allowed to change there.

   • Drift checks across strategy, decisions, artifacts, and roadmap work
   • Human review before correcting material company state
   • Clear difference between proposed and approved direction
   • Restore selected workspace objects to last approved stateComing soon
   • External rollback only where the connected system supports it