SECURITY · AI AGENTS
AI agent governance at Stacksync.
Stacksync AI Copilot and Genies run on the same platform as your two-way sync and workflows, under the same role-based access, approval steps and log explorer. You decide which writes wait for a person, trace each run in the logs, and can build everything by hand when a policy keeps AI out of scope.
AT A GLANCE
AI governance controls, answered in one table
| Reviewer question | Answer | Verify on |
|---|---|---|
| What are the AI features called? | Stacksync AI Copilot builds integrations and workflows from a prompt. Genies are AI agents that run tasks across your systems. Neither is related to Microsoft Copilot. | AI agents |
| Can a person approve a write first? | Yes. An approval step pauses the run with the full payload until someone approves, rejects or edits it. | Workflow automation |
| Who can build, run and deploy? | Role-based access control on every plan. SSO and SCIM on the Enterprise plan. | Pricing |
| Can we trace what an agent did? | The log explorer shows each run and its steps on every plan. Audit logs are listed on the Enterprise plan. | Pricing |
| Are dev and production separate? | Enterprise includes separate Dev, Staging and Production environments. Other plans include one. | Pricing |
| Can we run with AI out of scope? | Yes. Build two-way sync and workflows by hand in no-code or pro-code. | Two-way sync |
| Which certifications apply? | SOC 2 Type II and ISO 27001, and HIPAA under a BAA, listed on the Pro plan and up. Confirm AI feature scope in your review. | Compliance |
| Which models run, and where? | Request the model providers, inference region, retention and training terms in writing during security review, and check them against the DPA and the sub-processor list. | Trust Center |
Three names
Stacksync AI Copilot, Genies and Microsoft Copilot are three different things
Security questionnaires often mix these up, so start with the names.
- Stacksync AI Copilot is the builder. You describe an integration or workflow in a prompt, and it builds it in the Stacksync workflow builder, where your team can open it, edit it in no-code or pro-code and version it.
- Genies are Stacksync AI agents. They read your connected systems and act across them, running workflows, two-way sync and EDI tasks. See AI agents.
- Microsoft Copilot is a Microsoft product. It has no link to Stacksync AI Copilot or Genies. Stacksync reaches Dynamics 365 and other Microsoft apps through connectors, the same way it reaches NetSuite or Salesforce.
A fourth path shows up in reviews too. If you connect your own assistant, such as Claude, ChatGPT or Gemini, to your synced data, for example through a Postgres connection or MCP, that assistant's provider processes your prompts and whatever the assistant reads back, under your company's agreement with that provider. Govern it the way you govern that assistant today.
Approvals and access
Hold AI agent writes until a person approves them
Put an approval step in front of any write you want a person to see first. The run pauses with the full payload and waits. The reviewer approves it, rejects it or edits it, and the run continues from where it stopped. Stacksync logs each decision with the person who made it.
Example: a new NetSuite customer, created only after approval
- A Genie reads a closed-won deal in your CRM and drafts the NetSuite customer record.
- An approval step holds the draft for a finance reviewer.
- On approval, the workflow creates the customer in NetSuite.
- A Slack step tells the sales team the account exists.
Set the rules around the gate yourself. For vendor bills, a condition before the approval step can route only bills above an amount you choose to a reviewer and let smaller ones post. That threshold lives in your workflow logic; Stacksync has no global approval-threshold setting.
Who can build, run and deploy
Role-based access control on every plan decides who can build, run and deploy workflows. SSO and SCIM come with the Enterprise plan. The Enterprise plan adds separate Dev, Staging and Production environments, so you can test a Genie against sandbox data before anything reaches production. Compare plans on pricing.
For more on where to place a human step, read when to escalate automation to a human.
Audit
Trace what an AI agent did, and prove it ran once
Every plan includes the log explorer. Open a run to see its steps and where it failed, then replay the failed run once you fix the cause. Stacksync logs each approval decision and who made it.
- Audit logs. The pricing compare table lists audit logs of user and configuration activity on the Enterprise plan.
- Retention. How long Stacksync keeps logs depends on your plan. Enterprise can store logs in your own storage, under your own retention policy.
- Alerts. Every plan alerts by email. Enterprise adds Slack, PagerDuty, webhooks, Sentry, Datadog and other channels.
Checking for duplicates and gaps
Logs show what ran. To show that a NetSuite vendor bill posted once and matches its source, build the check in:
- Before the write, look the bill up in NetSuite by vendor and invoice number. If it exists, stop the run or send it to approval.
- On a schedule, have a second workflow or a Genie compare posted bills with the source records and flag duplicates, missing bills and mismatched amounts.
- Send the flags to the reviewer who owns the process, through Slack or email.
This monitoring is a pattern you build with the same tools. Stacksync does not sell it as a prebuilt agent.
Data flow and compliance
What the certifications cover, and what to get in writing for AI
Stacksync holds SOC 2 Type II and ISO 27001. Stacksync supports HIPAA-aligned deployments under a Business Associate Agreement (BAA). The pricing page lists these frameworks on the Pro plan and up; GDPR and CCPA apply on every plan. Encryption and processing regions are described on Security, and each framework has a page under Compliance, including HIPAA.
AI features raise questions a general platform review does not. Ask the Stacksync security team to answer these in writing, and check the answers against the Data Processing Addendum and the sub-processor list in the Trust Center:
- Models and subprocessors. Which model providers power Stacksync AI Copilot and Genies, and whether they appear on the subprocessor list.
- Inference region. Where AI inference runs, and whether it stays inside the processing region you select. The processing region on Security covers sync processing; the page does not yet state where AI inference runs.
- Retention. What Stacksync keeps from prompts, Copilot chats and agent runs, for how long, and whether you can export it.
- Training. Whether any provider may train on your data, and how a Genie using your workspace context or your feedback differs from model training.
- Scope. Whether the AI features you plan to use sit inside the SOC 2 report's system description and your BAA.
Buyers we spoke with in healthcare, SEC-regulated finance and higher education raised most of these before approving AI on patient, financial or student data. Bring the list to your first call with Stacksync.
AI out of scope
Run two-way sync and workflows with AI features out of scope
Some boards and IT teams bar financial or personal data from AI tools. You can still use Stacksync. Build two-way sync and workflows by hand in the no-code builder or in pro-code, without the AI Copilot or Genies. Two-way sync follows the field mappings and write directions you configure.
Tell us about the policy during scoping, and ask for the proof of concept to run without AI features. Point it at a sandbox copy of your system if production data must stay out too.
PII and PHI
Keep sensitive fields away from AI agents
- Leave fields out. Map only the fields a sync or workflow needs. SSNs, diagnosis codes and bank details that stay out of the mapping never reach a Genie.
- Mask in code. Stacksync has no built-in masking setting. To mask or hash a value, transform it in a pro-code step before it moves on.
- Limit who builds. Role-based access control decides who can build, run and deploy workflows that touch sensitive systems.
Replacing n8n, Zapier, or staff pasting records into a chat assistant? Start here: sync the regulated system, exclude the sensitive fields, and give agents the rest.
FAQ
AI agent governance: questions from security reviews
Is Stacksync AI Copilot the same as Microsoft Copilot?
No. Stacksync AI Copilot is the builder inside Stacksync: you describe an integration or workflow in a prompt and it builds it in the Stacksync workflow builder. Genies are Stacksync AI agents that run tasks across your connected systems. Microsoft Copilot is a separate Microsoft product with no link to either. Stacksync reaches Microsoft apps such as Dynamics 365 through connectors, like any other system.
Does our data go to OpenAI or Anthropic when we use AI in Stacksync?
It depends on the path. If you connect your own assistant, such as Claude, ChatGPT or Gemini, to your synced data, for example through a Postgres connection or MCP, that provider processes your prompts and whatever the assistant reads back, under your agreement with that provider. For Stacksync AI Copilot and Genies, ask the Stacksync security team for the model providers, the region where inference runs, and the retention and training terms in writing, and check them against the Stacksync Data Processing Addendum and the sub-processor list in the Stacksync Trust Center.
How do I make an AI agent wait for a human to approve a NetSuite write?
Add an approval step to the workflow before the write. The run pauses with the full payload and waits for a person to approve, reject or edit it, then continues. For example, a Genie drafts a new customer record, the workflow holds it for a finance reviewer, and after approval it creates the record in NetSuite and posts a note to the sales team in Slack. Each approval is logged and attributable.
Can we see every action an AI agent took?
Every plan includes the log explorer, which shows each workflow run and its steps, and you can replay a failed run. /pricing lists audit logs of user and configuration activity on the Enterprise plan. Log retention depends on your plan, and Enterprise can store logs in your own storage.
How do I check that an agent did not create a duplicate NetSuite vendor bill?
Build the check into the workflow. Before the write, look up the bill in NetSuite by vendor and invoice number and stop or route to approval when a match exists. After the fact, a scheduled workflow or a Genie can compare posted bills against the source records and flag duplicates or gaps. This is a pattern you build with the same tools; Stacksync does not ship a prebuilt monitoring agent for it.
Can we use Stacksync with AI features out of scope?
Yes. You can build two-way sync and workflows by hand in the no-code builder or in pro-code, without the AI Copilot or Genies. Two-way sync follows the field mappings and write directions you configure. If a board or IT policy keeps AI out, say so during scoping so the proof of concept runs without AI features, for example on a sandbox copy of the system.
Are the AI Copilot and Genies covered by SOC 2 and HIPAA?
Stacksync has not yet published whether the AI Copilot and Genies sit inside its SOC 2 report scope and BAA, so confirm it in writing during your security review. Stacksync holds SOC 2 Type II and ISO 27001. Stacksync supports HIPAA-aligned deployments under a Business Associate Agreement (BAA). /pricing lists these on the Pro plan and up. Request the current SOC 2 report and read its system description to see which features it covers.
How do we keep SSNs and other PII away from AI agents?
Leave sensitive fields out of the field mapping and out of the steps a workflow or Genie reads. Stacksync has no built-in masking setting; to mask or hash a value, transform it in a pro-code step before it moves. Role-based access control decides who can build, run and deploy workflows, on every plan.
