Enterprise Work Platforms: AI Governance and Data Privacy
The short answer
All five platforms promise that their AI respects your existing permissions. They differ on what happens to your data afterwards. As of October 27, 2026:
- Microsoft 365 (Microsoft Copilot) and Google Workspace (Gemini) — your prompts and content are not used to train foundation models; existing DLP, labels, audit and eDiscovery apply.
- Salesforce (Agentforce, Einstein Trust Layer) — the most explicit controls for customer data: masking, zero data retention with external model providers, toxicity checks and an audit trail.
- ServiceNow (Now Assist, AI Control Tower) — PII masking before prompts reach the model, an opt-out from data sharing, and an inventory of every AI agent and model.
- Atlassian (Rovo) — permission-aware, but since August 17, 2026 it contributes de-identified customer metadata to improve AI for all customers; only Enterprise plans can switch that off.
This page is a spoke of our trusted agent control plane providers guide, which covers governing agents across vendors.
The comparison
| Platform | AI product | Trains on your data? | Privacy controls | Governance and audit |
|---|---|---|---|---|
| Microsoft 365 | Microsoft Copilot, Copilot Chat | No — prompts, responses and Graph data not used to train foundation models | Enterprise data protection, Purview sensitivity labels and DLP | Purview audit, eDiscovery, retention policies |
| Google Workspace | Gemini in Workspace apps and Gemini app | No — not used for model training outside your domain without permission; no human review | Workspace DLP, client-side encryption, per-OU feature controls | Audit logs, Reports API, Vault for Gemini conversations |
| Salesforce | Agentforce | No — zero data retention agreements with external LLM providers | Einstein Trust Layer: secure retrieval, dynamic grounding, data masking, toxicity detection | Audit trail of prompt → response; logs to your own Data Cloud |
| ServiceNow | Now Assist, AI agents | Opt-out of data sharing available | Now Assist Data Privacy masks PII with placeholders before the LLM call | AI Control Tower: discovery, inventory, risk and compliance across in-house and third-party AI |
| Atlassian | Rovo, Rovo Dev | De-identified metadata contributed by default since Aug 17, 2026; in-app data on Free/Standard | Permission sync with connected apps; data residency | Admin controls, audit logs; data contribution settings in Atlassian Administration |
Microsoft and Google: the productivity suites
Both suites make the same core promise and back it with tools you already run. Microsoft’s documentation for enterprise data protection (Microsoft 365 Copilot was renamed Microsoft Copilot in 2026) states that prompts, responses and data accessed through Microsoft Graph are not used to train foundation models, and Copilot only shows content the user can already open. Google’s Workspace privacy hub says Gemini does not use your prompts, Workspace content or generated responses to train generative AI models without permission, and that content is not human-reviewed. Google adds client-side encryption, which makes protected files unreadable to Gemini entirely.
The practical risk in both is over-sharing, not training. If a sensitive file is shared with “everyone in the company”, the assistant will find it. Fix permissions and labels before rollout.
Salesforce and ServiceNow: systems of record
These platforms hold customer and case data, so their controls focus on what leaves the platform. Salesforce’s Einstein Trust Layer masks sensitive fields before a prompt is sent, relies on zero data retention agreements so model providers do not store prompts or responses, scans for toxic output and keeps an audit trail. Agentforce is priced separately — $2 per conversation, $500 per 100,000 Flex Credits, or $125 per user per month for the unmetered employee add-on (Salesforce list prices, October 27, 2026).
ServiceNow’s Now Assist Data Privacy replaces personal data with placeholders before the model call and restores it in the response; admins configure the policies and can opt out of data sharing. The AI Control Tower inventories agents and models — including third-party ones — which is what most auditors ask for first.
Atlassian: read the data contribution settings
Atlassian changed its data practices on August 17, 2026. Defaults follow the highest plan in your organisation: Free and Standard contribute metadata and in-app data; Premium and Enterprise contribute metadata but not in-app data. Metadata (content attributes and common patterns from search and Rovo Chat) is de-identified and aggregated, but only Enterprise customers can opt out of it. If your compliance policy forbids any cross-customer use, you need the Enterprise plan or one of the compliance exclusions Atlassian lists.
How to choose
- Employee productivity, existing Microsoft estate: Microsoft Copilot; your Purview policies carry over.
- Employee productivity, Google estate: Gemini in Workspace; turn on client-side encryption for the most sensitive drives.
- Customer data in AI workflows: Salesforce or ServiceNow, depending on which is your system of record.
- Engineering knowledge on Jira and Confluence: Rovo, but check data contribution settings in Atlassian Administration and budget for Enterprise if you need a full opt-out.
- Agents spanning all of the above: a cross-vendor layer; see best AI agent control planes.
Last verified: October 27, 2026.