AI & personal dataUpdated 22 November 20269 min read

Enterprise AI Copilots and Employee Data Under DPDP

How the section 7(i) employment legitimate use, Rule 6 safeguards and Rule 8(3) retention apply to enterprise copilots that read email, documents, chats and code.

By Suhas Bhairav · Reviewed against the DPDP Rules, 2025 (G.S.R. 846(E), 13 Nov 2025)

Employee data basis
s. 7(i) employment legitimate use (or consent)
Customer data in copilots
Original lawful basis and purpose apply
Security safeguards
Rule 6(1)(a)–(g)
Interaction log retention
Minimum one year, then erase (Rule 8(3))
Rules apply from
13 May 2027

Quick answer

Enterprise AI copilots process employees' personal data and, through email and documents, customers' and third parties' data too. Employee data may rely on the section 7(i) employment legitimate use, limited to employment-related purposes, but customer data still needs its own basis. Rule 6 safeguards and Rule 8(3) one-year retention apply from 13 May 2027, and permissions oversharing is the biggest practical risk.

  • Section 7(i) covers processing for employment purposes, not every analytic an employer might run on copilot telemetry.
  • Copilots inherit document permissions; overshared drives become instant data leaks through natural-language search.
  • Customer and third-party data inside email and files is processed under the basis on which it was collected, not under s. 7(i).
  • Prompt logs and copilot interactions are processing logs subject to Rule 8(3) one-year retention, then erasure.
  • If the copilot vendor is a processor, section 8(2) requires a valid contract and Rule 6(1)(f) security clauses.

Who this is for: CIOs and IT leaders, Security and identity teams, HR and employee relations, Data Protection Officers, Global capability centres rolling out copilots

What personal data does an enterprise copilot process?

An enterprise copilot — an assistant embedded in email, documents, chat, meetings, CRM or code tools — processes three groups of personal data at once:

  1. Employees' own data. Their emails, calendars, chats, documents, meeting transcripts, and the prompts they type.
  2. Colleagues' data. Everything about other employees that appears in the content the copilot can read: performance discussions, salary sheets, medical leave emails.
  3. Customers' and third parties' data. CRM records, support tickets, contracts and email threads with external people.

Plus a fourth category the copilot creates: interaction telemetry — who asked what, which files were retrieved, usage frequency, and sometimes productivity metrics.

Under the DPDP Act the employer is the Data Fiduciary for all of this. The copilot vendor is usually a Data Processor, which means section 8(1) keeps the employer responsible for the vendor's processing and section 8(2) requires a valid contract.

Each category needs its own analysis because the lawful basis differs. Getting this right before 13 May 2027 is mostly about access control and purpose limits, not new consent screens.

How far does the employment legitimate use stretch?

Section 7(i) of the DPDP Act allows processing without consent for the purposes of employment or those related to safeguarding the employer from loss or liability, such as prevention of corporate espionage, maintenance of confidentiality of trade secrets, intellectual property and classified information, or providing a service or benefit sought by an employee.

For copilots this supports:

  • Giving employees an AI assistant to do their jobs.
  • Processing employees' work content through the copilot.
  • Security monitoring of copilot use to prevent data exfiltration or leakage of confidential information.

It is a weaker fit for:

  • Productivity scoring of individual employees from copilot telemetry, used for performance decisions. It may be arguable as "employment purposes", but it is the kind of use where necessity and proportionality will be questioned.
  • Sentiment or wellbeing analysis of employees' messages.
  • Reuse for unrelated purposes, such as training a vendor's general model on employee content.

Section 7(i) also does not cover customers' data inside email and files. That data remains governed by the basis under which it was collected — often consent or section 7(a) — and the copilot should use it only within that purpose.

The HR-side analysis is in DPDP for HR and employee data.

The real risk: permissions oversharing

Copilots retrieve whatever the signed-in user can access. Most organisations have years of overshared folders — "anyone in the company with the link", stale team sites, HR exports in shared drives. Before copilots, that content was technically accessible but practically hidden. A natural-language query such as "show me salary details for the sales team" makes it instantly discoverable.

Under Rule 6(1)(b) — appropriate measures to control access to computer resources — this is a security safeguard failure, and an employee retrieving data they should not see can amount to a personal data breach requiring Rule 7 notice.

Remediation before broad rollout:

  1. Discover. Scan for files with organisation-wide or public link sharing that contain personal data patterns (PAN, salary, medical, Aadhaar-format numbers).
  2. Label. Apply sensitivity labels to HR, finance, legal and health content.
  3. Exclude. Remove highly sensitive sites and mailboxes from copilot retrieval scope.
  4. Restrict. Convert "anyone with the link" sharing to named groups.
  5. Pilot narrowly. Start with teams whose data access is already well scoped.
  6. Monitor. Alert on copilot queries that retrieve labelled sensitive content.

Copilot logs, retention and employee rights

Copilot interaction logs — prompts, retrieved file references, responses — are processing logs.

  • Retention. Rule 8(3) requires personal data, traffic data and processing logs to be retained for at least one year from processing for Seventh Schedule purposes, then erased unless another law requires longer. Rule 6(1)(e) requires security logs for one year. Configure the copilot platform's audit retention accordingly, and ensure it erases after the period.
  • Content vs metadata. Keep retrieval references and event metadata for the year; limit retention of full prompt and response text where the platform allows. See LLM logging under the one-year rule.
  • Access to logs. Restrict who can search employees' copilot history. HR investigations should follow a documented process with approval, not ad hoc searches by administrators.
  • Departing employees. Their mailbox and files are often retained for business reasons; their copilot interaction history should follow the same retention schedule, then be erased.
  • CERT-In. Copilot audit logs are ICT logs; keep them for the rolling 180 days within India under the April 2022 Directions.

Employees remain Data Principals. Publish a contact for questions about processing (Rule 9) and a grievance channel (Rule 14(3)) that covers employee concerns.

Vendor terms and cross-border processing

Most enterprise copilots run on a global cloud vendor's infrastructure. Check:

  • Training use. Enterprise terms should state that prompts, responses and organisational content are not used to train the vendor's foundation models.
  • Processing location. DPDP does not generally require localisation; section 16 restricts transfers only to countries the Government notifies, and Rule 15 lets the Government specify requirements for making data available to foreign States. Sectoral rules — RBI payment data storage, for example — may be stricter and continue to apply.
  • Sub-processors and model providers. Copilots often call a separate model provider; confirm it is covered by the same terms.
  • Plugins and connectors. Third-party plugins can send data to additional vendors outside your main contract. Approve them centrally.
  • Breach notice. Vendor commitments must let you meet Rule 7: notice to affected people without delay and a detailed Board report within 72 hours.

For Indian GCCs serving European parents, employee data may also be subject to GDPR, and AI systems used for monitoring and evaluating workers' performance or behaviour fall within Annex III high-risk employment uses under the EU AI Act. See EU AI Act high-risk HR rules and the AI vendor due diligence checklist.

An acceptable-use policy that actually works

A copilot policy should be short enough that employees read it.

  • Approved tools only. Use the enterprise copilot; do not paste customer or employee data into consumer AI tools.
  • No sensitive categories in prompts unless the workflow is approved: health, financial account details, children's data, identity numbers.
  • Check before sharing. Copilot-generated summaries can include data from documents the recipient should not see.
  • No covert monitoring claims. Tell employees what copilot telemetry is collected and how it is used; this is good practice under section 7(i) even though a consent notice is not required for legitimate uses.
  • Report mistakes. Make it easy to report accidental exposure, because Rule 7 notification starts when the organisation becomes aware.

Pair the policy with training: the MeitY AI Governance Guidelines emphasise capacity building, and if you serve the EU, AI literacy obligations under the EU AI Act have applied since 2 February 2025.

Further reading on suhasbhairav.com

Before 13 May 2027

Compliance checklist

  1. Map data categories the copilot can reach

    List employee, colleague, customer and third-party data across mail, files, chat and CRM.

    DPDP Act s. 8(1)

  2. Document the lawful basis per category

    Rely on s. 7(i) for employment purposes and the original basis for customer data.

    DPDP Act s. 7(i)

  3. Fix overshared content before rollout

    Remove organisation-wide links on sensitive files and exclude HR and finance sites from retrieval.

    Rule 6(1)(b)

  4. Apply sensitivity labels

    Label HR, health, finance and legal content and alert on copilot retrieval of labelled items.

    Rule 6(1)(c)

  5. Configure one-year audit log retention

    Keep copilot interaction logs for at least one year and erase after.

    Rule 8(3)

  6. Confirm no-training terms with the vendor

    Get written terms that prompts and content are not used to train foundation models.

    DPDP Act s. 8(2)

  7. Govern plugins and connectors

    Approve third-party plugins centrally and treat each as a processor.

    DPDP Act s. 8(2)

  8. Limit productivity analytics

    Avoid individual performance scoring from copilot telemetry unless necessity is documented.

    DPDP Act s. 7(i)

  9. Publish employee contact and grievance channel

    Give employees a route to ask about and complain about copilot processing.

    Rules 9 and 14(3)

Key dates

  1. EU AI Act AI literacy obligations apply (EU-facing teams)

  2. DPDP Rules notified

  3. Rules 6, 7, 8(3), 9 and 14 apply to copilot processing

Answers

Frequently asked questions

Do I need employee consent to deploy an AI copilot in India?

Generally not for employment purposes. Section 7(i) of the DPDP Act allows processing for employment and for safeguarding the employer against loss or liability without consent. Customer and third-party data the copilot reads is a separate question, governed by the basis on which that data was originally collected.

Can I use copilot telemetry to measure employee productivity?

It is arguable under the employment legitimate use, but it is where necessity and proportionality are most likely to be challenged. Aggregate adoption metrics are low risk; individual scoring used for performance decisions is higher risk. Tell employees what is collected, limit access and document why it is necessary.

Is it a data breach if a copilot shows an employee someone else's salary?

It can be. Under the DPDP Act a personal data breach includes unauthorised access or disclosure. If a copilot surfaces data the user should not have seen because of overshared permissions, assess it as a breach and, where it is one, notify affected individuals and the Board under Rule 7.

How long should copilot prompt logs be kept?

From 13 May 2027, Rule 8(3) requires processing logs to be kept for at least one year from processing and then erased unless another law requires longer. CERT-In separately expects ICT logs for 180 days within India. Keep metadata and retrieval references for the year and limit full-text retention.

Does section 7(i) cover candidates and contractors using copilots?

Section 7(i) refers to purposes of employment. Whether it extends to job applicants or independent contractors is not spelled out, so take a cautious approach: give them a clear notice, limit copilot access to what their role requires, and consider consent or contract-based terms where employment is doubtful.

Do EU rules apply to Indian GCC employees using copilots?

GDPR applies where the processing falls within its territorial scope, for example data of EU employees or processing for an EU establishment. The EU AI Act lists AI used for monitoring and evaluating workers' performance or behaviour as Annex III high-risk, so check scope carefully if copilot analytics feed HR decisions.

Legal references and sources

  • Digital Personal Data Protection Act, 2023 — Section 7(i) (official text)
  • Digital Personal Data Protection Act, 2023 — Sections 8(1), 8(2) and 8(5) (official text)
  • Digital Personal Data Protection Rules, 2025 — Rule 6 and Rule 8(3) (official text)
  • EU AI Act (Regulation (EU) 2024/1689) — Annex III (employment) (official text)

This guide explains the law for engineering and business teams. It is not legal advice; confirm how it applies to your organisation with qualified counsel, and check the Gazette for later notifications or amendments.

About the author

Suhas Bhairav

Suhas Bhairav

Systems architect and applied AI practitioner who designs production AI systems for Indian and global companies, with a focus on privacy-by-design under the DPDP Act and the EU AI Act.