What Happens to Your Data When Staff Use Microsoft Copilot, Google Gemini, and ChatGPT

June 11, 2026 · 6min read · The Technica Stack

When a Philippine staff member types client data into a Microsoft Copilot prompt, submits a spreadsheet to Google Gemini, or pastes a confidential document into ChatGPT, two questions matter: does that data leave the organisation's control, and is it used to train the AI model?

The answers are different for each tool — and they depend on the specific plan tier and configuration. Getting this wrong has direct implications under RA 10173 (Data Privacy Act): if personal data is submitted to a tool that uses it for model training or exposes it beyond intended scope, the organisation may have a reportable personal data breach.

This article covers confirmed data handling practices for the three AI tools most Philippine offices are using in 2026.

Microsoft Copilot (in Microsoft 365)

Data Processing

Microsoft 365 Copilot processes data entirely within your organisation's Microsoft 365 tenant. When a user asks Copilot a question, the query and any accessed document content are processed by Microsoft's AI models but the data flows through Microsoft's enterprise infrastructure — not shared with OpenAI's training systems or other customers.

Microsoft's commitment (confirmed, June 2026):

Your Microsoft 365 data is not used to train Microsoft's foundation AI models
Copilot does not store prompts or responses outside your tenant
All data processing occurs within your Microsoft 365 tenant boundary
For Philippine tenants: data residency is Singapore (Asia Pacific) region

What Copilot Can Access

Copilot accesses data through Microsoft Graph — the same data your user has permission to access in SharePoint, OneDrive, Exchange, and Teams. This is both the power and the risk:

Power: Copilot can search across all your organisation's documents to ground its responses in your actual business data.

Risk: If your SharePoint permissions are too broad — if documents that should be restricted are accessible to all users — Copilot will surface them to users who technically have access but shouldn't practically see them. This is a permissions governance issue, not an AI issue, but AI makes it visible.

The action required: Before enabling Copilot, audit SharePoint site permissions and OneDrive sharing settings. Documents that should not be generally accessible must be locked down before Copilot amplifies the exposure.

What Microsoft Staff Can See

Microsoft's contractual commitment (Microsoft Product Terms, DPA) prohibits Microsoft personnel from accessing customer content except for defined operational purposes (support incidents, security incidents) with logging and approval requirements.

Google Workspace Gemini

Data Processing and Training Commitment

Google's data processing commitment for Workspace varies by plan tier — this is the critical nuance most Philippine organisations miss.

Workspace Business Standard, Plus, Enterprise: Google does not use Workspace Customer Data (your emails, documents, chats, files) to train Google's AI models. The commitment is explicit in Google's Workspace data processing addendum.

Google One (personal) or free Google accounts: Google's terms allow use of content to improve products, including AI. This includes Gemini used through personal Gmail accounts.

The Philippine office risk: Staff using personal Gmail or Google One accounts for work — or using Gemini through personal accounts rather than the organisation's Workspace account — are on the personal-account terms, not the enterprise terms. Their inputs may be used for training.

The action required: Ensure all staff access Gemini through the organisation's Google Workspace account, not personal Google accounts. Enable Context-Aware Access to block access from personal accounts to Workspace resources.

Data Residency

For Philippine Google Workspace tenants, data processing occurs in Google's Singapore-region infrastructure for most Workspace services. Confirm your specific region setting in the Google Admin console.

Gemini Interaction Logs

Workspace admins can access Gemini interaction logs through Google Vault (if configured). This means your IT or compliance team can audit what staff are asking Gemini — relevant for compliance with RA 10173's purpose limitation principles.

ChatGPT (OpenAI) — Consumer and Enterprise

This is where the distinction matters most and where Philippine organisations are most exposed.

ChatGPT Free and Plus (Personal)

OpenAI's data handling (free and Plus accounts):

Conversations may be used to train OpenAI's AI models by default
Users can opt out of training data use in settings, but many do not
Conversation data is stored on OpenAI's servers (US-based) indefinitely unless deleted

If a Philippine staff member pastes client data, employee records, financial information, or any personal data into ChatGPT Free or Plus, that data has potentially left the organisation's control and may be used for model training. This is a reportable data breach scenario under RA 10173 if personal data is involved.

ChatGPT Team

OpenAI's Team plan includes a commitment that conversations are not used for model training by default. Conversations are retained for 30 days.

ChatGPT Enterprise

OpenAI's Enterprise plan provides:

No training on customer data
Data encryption at rest and in transit
SOC 2 Type II compliance
Custom data retention controls

OpenAI API (via Azure OpenAI or direct)

When accessing OpenAI models through the API (including Azure OpenAI Service), OpenAI does not use API inputs for training. The Azure OpenAI commitment is explicit: customer data submitted through Azure OpenAI is not used to train Microsoft or OpenAI foundation models.

The practical implication for Philippine offices: staff using ChatGPT.com in a browser (free or Plus) are on consumer terms. The appropriate path for business use is either Azure OpenAI Service (enterprise, via Azure) or ChatGPT Enterprise.

The NPC Compliance Lens

Under RA 10173, personal data that is accidentally or negligently exposed to an AI model that uses it for training may constitute a personal data breach. The NPC's notification requirements apply when the breach "is likely to give rise to a real risk of serious harm."

What this means practically:

A staff member pasting a list of client names, SSS numbers, or phone numbers into ChatGPT Free → potential reportable breach
A staff member drafting a client email in Copilot within the M365 tenant → not a breach (data stays within tenant)
A staff member submitting employee salary data to Gemini through a personal Google account → potential breach

The risk is not hypothetical — it is occurring daily in Philippine offices that have not established an AI acceptable use policy and data input controls.

What Philippine Organisations Should Do

Classify which AI tools are approved for business use — tenant-managed tools (M365 Copilot, Workspace Gemini) vs consumer tools (ChatGPT Free, Claude.ai personal)
Define data input rules — no personal data (SSS, PhilHealth, names+IDs), financial data, or client confidential data in consumer AI tools
Ensure staff use organisational accounts — Gemini through Workspace account, not personal Gmail; Copilot through M365 tenant, not personal Microsoft account
Configure data governance controls — SharePoint permissions audit before Copilot; Context-Aware Access for Workspace before Gemini rollout
Write an AI Acceptable Use Policy — see our guide at /insights/ai-acceptable-use-policy-philippines-2026

For Philippine organisations establishing AI governance or deploying Microsoft 365 Copilot and Google Workspace Gemini safely, get in touch.

Talk to our Cloud & I.T. team →

Microsoft Copilot vs Google Gemini for Productivity: Which Works Better for Philippine Office Tasks?

11 June 2026

AI for Philippine HR Teams: What Entropy and AI Tools Actually Automate in Payroll and Compliance