DPDP Act 2023 + DPDP Rules 2025 — Complete AI & Generative AI Compliance Guide
Every Indian business using ChatGPT, Google Gemini, Microsoft Copilot, Anthropic Claude, Midjourney, or any other AI tool to process work — and that means almost every business in India right now — is engaging in an activity directly regulated by the Digital Personal Data Protection Act, 2023. When a lawyer drafts a contract using ChatGPT and pastes in a client's personal details, that is regulated data processing. When an HR team uses an AI recruitment tool to screen candidate CVs, that is regulated data processing. When a hospital uses an AI diagnostic tool trained on patient data, that is regulated data processing. The DPDPA applies to every AI tool your team uses — and almost no Indian business has yet built the compliance architecture to match. This is the only guide in India that covers AI and generative AI under DPDPA comprehensively — from ChatGPT's data practices to AI vendor DPAs to training AI on Indian customer data — all in one place.
100M+
Indian users of ChatGPT alone — plus millions more using Gemini, Copilot, Claude, and other AI tools daily
₹250 Cr
Maximum penalty if customer data input into an AI tool leads to a breach or unlawful disclosure
0
AI-specific exemptions in DPDPA — the Act applies to AI data processing exactly as it applies to every other form
May 2027
Full enforcement deadline — every AI use case in your business must be compliance-reviewed before then
Nov 2025
DPDP Rules notified — Board operational
Now
AI tools already in use — compliance obligations already active
May 2027
Full enforcement — all AI use cases must be compliant
📋 What This Guide Covers
- Why AI tools trigger DPDPA obligations — the legal framework
- The AI data flow map — what happens to data you put into AI tools
- ChatGPT and DPDPA — the specific compliance issues
- Google Gemini (Workspace AI) and DPDPA
- Microsoft Copilot and DPDPA
- Anthropic Claude and DPDPA
- Other AI tools — Midjourney, Perplexity, Grok, Grammarly, and more
- AI in HR — recruitment tools, performance analytics, productivity monitoring
- AI in healthcare — diagnostic AI, telemedicine, and medical AI tools
- AI in finance — credit scoring, fraud detection, and robo-advisory
- Building your own AI — training on Indian customer or employee data
- AI-generated content and personal data — photos, voice, deepfakes
- The AI vendor DPA problem — what you must negotiate
- Employee AI usage policies — your internal governance obligation
- Consent for AI processing — what must be disclosed to your users
- The most common AI-DPDPA violations in Indian businesses today
- Practical AI compliance framework — 30-day implementation plan
1. Why AI Tools Trigger DPDPA Obligations — The Legal Framework
The DPDPA, 2023 was enacted before the current generative AI explosion — but its drafters were forward-looking enough that its provisions apply cleanly to AI data processing. Understanding why requires understanding what the Act regulates.
⚖️ The Four DPDPA Triggers Activated by Almost Every AI Tool Use
1
Data Collection — Section 2(t)
When you type a customer's name, email, medical condition, or financial situation into an AI tool, you are inputting "personal data" — data about an identifiable individual — into a third-party system. This is a collection and transfer of personal data to that system's operator.
2
Processing — Section 2(x)
"Processing" includes any operation on personal data — collection, storage, use, analysis, or sharing. An AI tool reading a document, analysing it, storing it in conversation history, and using it to generate a response processes that data across multiple of these categories simultaneously.
3
Third-Party Transfer — Section 9 + Section 16
Every major AI tool (ChatGPT, Gemini, Claude, Copilot) is operated by a foreign company — OpenAI (USA), Google (USA), Anthropic (USA), Microsoft (USA). Inputting personal data into any of these tools is an immediate cross-border transfer of personal data outside India. This simultaneously triggers the Data Processor DPA requirement and the cross-border transfer framework.
4
Purpose Limitation — Section 4(1)(b)
Personal data collected from your customers for one purpose (say, to provide a service) cannot be used for a different purpose (training an AI model, personalising AI responses for a different user, or sharing with an AI tool's operator for their commercial purposes) without fresh consent. This is the most frequently violated principle in AI data use.
The core question every Indian business must answer for each AI tool: What personal data is going in? Whose data is it? What lawful basis exists for sending it to this AI tool? Has the individual whose data it is been informed that their data may be processed by this AI system? And what does the AI provider's privacy policy say about how they use and retain what they receive? Most businesses today cannot answer a single one of these questions. That is the compliance gap.
2. The AI Data Flow Map — What Actually Happens to Data You Input
Before assessing compliance obligations for any specific AI tool, you must understand what technically happens to the data you input. The answer is more complex — and more legally significant — than most users realise.
| Data Destination |
What the AI Does With Your Input |
DPDPA Risk |
| Immediate processing |
Your input is sent to the AI provider's servers (outside India for all major tools) where it is processed by the AI model to generate a response. This processing happens at the moment of every prompt. |
HIGH — always |
| Conversation history |
Most AI tools retain conversation history — your prompts and the AI's responses — in their systems for a defined period. This means personal data you input remains on the AI provider's servers after your session ends. The retention period varies: ChatGPT retains by default; enterprise versions may offer shorter retention or deletion controls. |
VERY HIGH |
| Training data |
By default, consumer versions of most AI tools (ChatGPT Free/Plus, Gemini consumer, Grammarly consumer) may use your inputs to train future AI models. This is the most serious data protection risk: personal data you input about a customer could be incorporated into an AI model and potentially surface in responses to other users. Enterprise/API versions typically offer data-not-for-training settings — but defaults must be actively changed. |
CRITICAL |
| Third-party plug-ins / extensions |
ChatGPT plug-ins, Copilot extensions, and Gemini integrations with third-party apps create additional data sharing flows. When an employee uses a ChatGPT plug-in to search the internet or connect to a business tool, their prompt data (including any personal data in it) may be shared with that plug-in's operator as well as OpenAI. |
VERY HIGH |
| AI provider's commercial use |
Some AI providers' terms permit them to use input data for their own commercial purposes beyond model training — including safety research, product improvement, and regulatory compliance. Each provider's data practices differ materially and must be reviewed against your DPDPA obligations. |
MEDIUM–HIGH |
3–6. The Four Major AI Tools — DPDPA Assessment
GP
ChatGPT (OpenAI) — Free, Plus, and Enterprise
Operator: OpenAI, Inc. (San Francisco, USA) · Servers: USA primarily · Training opt-out: Available on Enterprise only
DPDPA Risk: HIGH
❌ DPDPA Problems with Consumer ChatGPT (Free/Plus)
- Training by default: OpenAI's consumer terms allow using conversations to train future models. Any customer personal data in prompts may be used for training.
- Conversation retention: Conversations retained for 30 days after deletion request in consumer versions — personal data does not immediately disappear.
- No DPA available: Consumer ChatGPT has no Data Processing Agreement available — it is a consumer product, not a B2B service.
- US data residency: All data processed on US servers — cross-border transfer with no India-specific safeguards.
- No access controls: No ability to restrict which employees use it, what data they input, or what settings they use.
✅ Compliant Path: ChatGPT Enterprise / API
- No training on your data: Enterprise terms explicitly state inputs are not used for model training.
- DPA available: OpenAI provides a Data Processing Agreement for Enterprise customers — review against DPDPA Section 9 requirements.
- Conversation deletion: Enterprise allows shorter retention periods and API-level deletion controls.
- Admin controls: Domain-level deployment with admin visibility over usage.
- Still cross-border: Even Enterprise processes data on US servers — cross-border transfer disclosure and DPA review still required.
DPDPA verdict: Using consumer ChatGPT (Free or Plus) to process any personal data of Indian customers or employees is a direct DPDPA violation — no lawful basis, no notice, no DPA, no cross-border compliance. Switch to ChatGPT Enterprise or OpenAI API with a reviewed DPA, or prohibit personal data input into consumer ChatGPT through your AI usage policy.
G
Google Gemini — Consumer and Google Workspace
Operator: Google LLC (Mountain View, USA) · Servers: USA and global · Training: Consumer yes, Workspace Admins can control
DPDPA Risk: HIGH
🔍 Consumer Gemini — Same risks as consumer ChatGPT
- Google may use consumer Gemini conversations to improve its AI models — includes any personal data in prompts
- No DPA for consumer use — no B2B data processing terms available
- Conversation history stored and accessible to Google's human reviewers in some configurations
🏢 Google Workspace with Gemini — Better but still requires work
- Google's Workspace DPA (GDPR-compliant) is available — requires DPDPA addendum review
- Workspace admins can turn off Gemini training on workspace data — must be actively configured
- Gemini's access to Gmail, Drive, and Calendar creates additional data flow concerns — each integration must be disclosed
- Data may still be processed on non-India Google servers — cross-border transfer disclosure required
DPDPA action required: If your organisation uses Google Workspace with Gemini enabled, your IT admin must: (1) verify Gemini training is disabled for your workspace, (2) review and update your Google Workspace DPA with a DPDPA addendum, (3) update your employee privacy notice and customer privacy policy to disclose Gemini processing, and (4) create an AI usage policy specifying what data employees may and may not input into Gemini.
M
Microsoft Copilot (M365 and Bing)
Operator: Microsoft Corporation (Redmond, USA) · Servers: USA and global Azure · GDPR DPA: Available with M365 enterprise
DPDPA Risk: MEDIUM-HIGH
Microsoft Copilot integrated into Microsoft 365 (Word, Excel, Outlook, Teams) has access to your organisation's entire Microsoft ecosystem — emails, documents, meetings, and chats — and uses this data to generate AI responses. This creates the most expansive AI data processing footprint of any AI tool, because Copilot can access and process content across your entire M365 environment simultaneously.
✅ Copilot M365 Advantages
- Microsoft commits not to use M365 data to train models
- GDPR-compliant DPA available for enterprise customers
- Data processed in Azure and stays within Microsoft's infrastructure
- Admin controls over what Copilot can access
❌ Remaining DPDPA Issues
- Microsoft's DPA is GDPR-compliant, not DPDPA-reviewed — addendum needed
- Cross-border transfer to US Azure — disclosure required in privacy notice
- Bing-connected Copilot (consumer) has weaker protections than M365 enterprise
- Customer data in Outlook/CRM integrated with Copilot creates additional consent questions
DPDPA action required: Review your Microsoft Enterprise Agreement and DPA. Add a DPDPA-specific addendum. Update your privacy notice to disclose Microsoft Copilot as a data processor. Create Copilot access policies that restrict what data employees can query through Copilot — particularly customer databases, employee HR records, and medical information.
A
Anthropic Claude (Consumer and API)
Operator: Anthropic, PBC (San Francisco, USA) · Servers: AWS (USA) · Training: Consumer default; API opt-out available
DPDPA Risk: MEDIUM-HIGH
Claude's consumer version (claude.ai) may use conversations for training — similar concerns to consumer ChatGPT. The Claude API and Claude for Enterprise offer stronger data protections: data is not used for training by default, a Data Processing Agreement is available, and businesses can request data deletion. Claude processes data on AWS infrastructure in the US — cross-border transfer disclosure required. For businesses building Claude-powered applications through the API, Anthropic's API DPA must be reviewed against DPDPA Section 9 requirements and supplemented with an India-specific addendum.
DPDPA action: Use Claude API or Claude for Enterprise, not consumer claude.ai, for any business use involving personal data. Obtain and review Anthropic's DPA. Add DPDPA addendum. Disclose in privacy notice.
7. Other AI Tools — The Long Tail of AI Compliance Risk
Beyond the four giants, Indian businesses use dozens of AI-powered tools — many of which process personal data with even weaker data protection than the major platforms. The compliance risk is not concentrated at ChatGPT — it is distributed across your entire AI software stack.
| AI Tool |
What It Does |
Personal Data Risk |
DPA Available |
DPDPA Risk |
| Grammarly |
AI writing assistant — reads and processes every document or email run through it |
Extremely high — reads full text of contracts, emails, HR letters, legal documents containing personal data of third parties |
Business DPA available |
HIGH |
| Midjourney |
AI image generation — prompts describing real people or scenarios may constitute personal data |
Medium — risk if prompts describe identifiable individuals (e.g. "generate a photo of [person] in [situation]") or if generated images are used for identity-related purposes |
Limited DPA |
MEDIUM |
| Perplexity AI |
AI search engine — queries may include personal data if employees search for customer information |
Medium-High — search queries containing personal data are sent to Perplexity's servers. Enterprise version has better controls. |
Enterprise only |
MEDIUM |
| Grok (X/Twitter) |
xAI's chatbot integrated into X platform — trained on public X content and user interactions |
High — Grok's training on public data raises questions about whether personal data on X (including Indian users' public posts) is used in training without consent |
Not available |
HIGH |
| AI Recruitment Tools |
HireVue, Eightfold, iCIMS, Workable AI — screen CVs, analyse interview videos, score candidates |
Critical — processes sensitive candidate data including biometric-adjacent video data, employment history, and AI-generated assessments that directly affect employment decisions |
Available — review needed |
CRITICAL |
| Salesforce Einstein AI |
AI features within Salesforce CRM — analyses customer data to generate predictions and recommendations |
High — processes customer personal data (full CRM records) to generate AI insights. Salesforce DPA available but requires DPDPA review and Einstein-specific addendum. |
DPA available |
MEDIUM-HIGH |
| Notion AI / Confluence AI |
AI features in knowledge management tools — analyses and generates content from workspace documents |
High — workspace documents frequently contain employee data, customer information, and confidential business data. AI access to the full workspace creates broad data processing exposure. |
Enterprise DPA available |
HIGH |
| DeepL / AI Translation |
AI translation — used to translate documents, emails, or customer communications |
High — translating customer contracts, medical records, or legal documents sends the full content (including all personal data) to the translation service's servers abroad |
DeepL Pro DPA available |
MEDIUM |
8–10. AI in HR, Healthcare, and Finance — Sector-Specific Obligations
👥
AI in HR — Recruitment Screening, Performance Analytics, and Employee Monitoring
AI tools in HR create unique DPDPA obligations because they make automated decisions that significantly affect individuals — hiring, promotion, termination — using personal data. Under DPDPA, this creates both the standard data processing obligations and additional transparency requirements for algorithmic decision-making (particularly for Significant Data Fiduciaries).
AI CV screening: Screening candidate CVs through AI without disclosing that AI is used in the selection process, and without specific candidate consent for CV processing by an AI system, violates both the notice obligation (Section 5) and the purpose limitation principle. The AI tool is a Data Processor — a DPA is required.
AI interview analysis: Video interview analysis tools (HireVue, interviewing.io) that use AI to analyse facial expressions, speech patterns, and responses collect biometric-adjacent data. Specific consent before the interview is mandatory. The candidate must know AI analysis is happening and what factors are assessed.
Productivity AI monitoring: AI tools that track keystrokes, screenshot frequency, mouse movement, or application usage time collect intimate behavioural data about employees. This must be disclosed in the employee privacy notice and monitoring policy before deployment. Covert installation is prohibited.
🏥
AI in Healthcare — Diagnostic AI, Telemedicine AI, and Medical Data Processing
Healthcare AI processes the most sensitive personal data — health conditions, diagnoses, treatment plans — and makes recommendations that directly affect patient welfare. DPDPA applies with full force, and the Board is expected to treat healthcare AI compliance as a priority enforcement area.
AI diagnostic tools: Using AI to analyse radiology images, pathology slides, or symptom data requires: patient consent for AI involvement in their diagnosis, disclosure of the AI tool used (including the vendor), and a DPA with the AI tool provider. Patients must be informed that AI is assisting in their care.
Using ChatGPT for clinical notes: A doctor typing patient symptoms, diagnosis, and treatment plan into ChatGPT to generate a discharge summary is sending highly sensitive health data to OpenAI's servers in the US — without the patient's knowledge or consent. This is a direct, serious DPDPA violation.
Training AI on patient data: Hospitals that train AI models using historical patient data must have: patient consent for that specific use (or a legitimate research use ground), data minimisation (de-identification where possible), and ethical committee approval alongside DPDPA compliance.
💰
AI in Finance — Credit Scoring, Fraud Detection, and Robo-Advisory
Financial AI creates automated decisions with the most direct and immediate impact on individuals — credit approval, loan rejection, fraud flagging. DPDPA, RBI guidelines, and SEBI frameworks apply simultaneously, creating the most complex AI compliance environment in any Indian sector.
AI credit scoring: Credit decisions must be explainable to the individual. Using AI to reject a loan application without being able to explain the AI's decision to the applicant — in plain language — violates both DPDPA transparency requirements and emerging RBI guidance on algorithmic lending.
Fraud detection AI: AI fraud systems that flag transactions and freeze accounts are making automated decisions with significant financial consequences. Customers must be informed that AI fraud detection is used, the criteria cannot be opaque, and there must be a human review process for disputes.
Robo-advisory: Automated investment advice using AI must disclose the AI nature of the advice, the data used to generate recommendations, and the limitations of algorithmic advice. RBI and SEBI have separate disclosure requirements that apply alongside DPDPA.
11. Building Your Own AI — Training on Indian Customer or Employee Data
An increasing number of Indian businesses are moving beyond using off-the-shelf AI tools to building their own — fine-tuning existing models, training custom models, or building RAG (Retrieval Augmented Generation) systems on proprietary data. This creates the most complex DPDPA compliance scenario in the AI landscape.
🧠
Training data consent — the most critical obligation
Every individual whose personal data is used to train an AI model must have consented specifically to that use — or you must have a legitimate use ground that extends to AI training (which is narrow and fact-specific). Using historical customer transaction data, support chat logs, or employee emails to train an internal AI model without specific consent for that purpose is a direct DPDPA violation. The fact that you originally collected the data for a different purpose (service delivery, customer support) does not authorise using it for AI training without fresh consent.
🔒
De-identification does not fully resolve the consent problem
Many companies attempt to sidestep the consent issue by de-identifying training data — removing names, emails, and obvious identifiers before training. Under DPDPA, de-identified data may still be personal data if the individual is re-identifiable by any means. True anonymisation — where re-identification is not reasonably possible even with additional information — takes the data out of DPDPA's scope. But de-identification that merely removes obvious identifiers while leaving re-identifiable patterns (unusual purchase combinations, unique medical histories, distinctive writing styles) is not anonymisation under the law. Get expert assessment before relying on de-identification as your compliance strategy.
🤝
Cloud AI training infrastructure — DPA required with cloud provider
If you are training a model on AWS SageMaker, Google Vertex AI, or Azure ML using Indian personal data, the cloud provider becomes your Data Processor for that training activity. Your existing cloud DPA must be reviewed to confirm it covers AI model training as a processing activity. If not, an addendum is required. Additionally, if your training data contains sensitive categories (health, financial, biometric), the security obligations on the training infrastructure are heightened accordingly.
📋
RAG systems — the "I'm just searching my own data" misunderstanding
Retrieval Augmented Generation (RAG) systems — where a company's proprietary documents are indexed in a vector database and retrieved by an AI to answer questions — are increasingly popular. Many companies believe RAG avoids training concerns because "the data isn't training the model, it's just being retrieved." This is a compliance misunderstanding. The personal data in your vector database is still being processed by the AI system every time a query is answered. The AI is still a Data Processor. The cross-border transfer obligations still apply if the RAG infrastructure is hosted abroad. And the privacy notice must disclose that an AI system has access to and processes personal data in your document store.
12. AI-Generated Content and Personal Data — Photos, Voice, Deepfakes
AI-generated content involving real people creates a distinct category of DPDPA risk that sits at the intersection of data protection and personality rights. This is an area where Indian law is still developing, but the DPDPA provides immediate protections.
🚫 Prohibited AI Content Uses
- Generating realistic images of identifiable real employees without consent — even for marketing materials
- AI voice cloning of employees or customers for any purpose without specific consent
- Using AI to generate content that uses a person's likeness in a false or misleading context
- Deepfake generation of any identifiable individual without explicit consent
- Using customer photos uploaded to a platform to train AI image models without consent
✅ Compliant AI Content Uses
- Generating AI avatars for employees who have specifically consented to avatar creation for defined purposes
- AI voice for text-to-speech of your own company content (no real person's voice)
- AI-generated illustrations that do not depict identifiable real individuals
- Using fully anonymised, non-identifiable images in AI training with documented anonymisation process
- Customer-initiated AI avatar creation where the customer is generating their own representation
13. The AI Vendor DPA Problem — What You Must Negotiate
The central contractual challenge in AI compliance is that most AI tools are designed as consumer products or developer APIs — their standard terms were not written with Indian data protection law in mind. Getting a DPDPA-compliant DPA from an AI vendor requires knowing what to ask for and what to push back on.
📋 What a DPDPA-Compliant AI Vendor DPA Must Include
①
No training on your data — explicit contractual commitment that input data and conversation history will not be used to train, fine-tune, or improve the AI model, and will not be used for any purpose other than generating the requested output.
②
Defined retention and deletion timelines — maximum retention periods for input data, conversation history, and any generated outputs, plus a contractual deletion obligation and confirmation mechanism.
③
Sub-processor restrictions and list — identification of all sub-processors (cloud providers, safety contractors, annotation vendors) with a requirement to notify you before adding new sub-processors and your right to object.
④
Security standards — minimum security requirements for data at rest and in transit, including encryption standards, access controls, and the AI vendor's security certification (SOC 2, ISO 27001).
⑤
Breach notification timeline — contractual requirement to notify you of any security incident affecting your data within a period that gives you sufficient time to meet your DPDPA Board notification obligation.
⑥
Data subject rights support — obligation on the AI vendor to assist you in responding to Data Principal rights requests (access, deletion) that require action on the vendor's systems.
⑦
Indian law and cross-border transfer compliance — acknowledgement that the processing is subject to Indian data protection law, and commitment to comply with any Board-notified restrictions on cross-border transfers, including a mechanism to stop transfers if your data's destination country is restricted.
⑧
Audit rights — your right to audit or receive third-party audit reports confirming the AI vendor's compliance with the DPA obligations.
14. Employee AI Usage Policy — Your Internal Governance Obligation
Under DPDPA Section 8(5), Data Fiduciaries must implement appropriate technical and organisational measures to ensure compliance. For AI tools, the most important organisational measure is a clear, enforceable employee AI usage policy. Without this, you cannot demonstrate that you have taken reasonable steps to prevent employees from inputting personal data into non-compliant AI systems.
📋 Employee AI Usage Policy — Minimum Required Content
What must be prohibited
- Inputting any customer personal data into consumer AI tools (ChatGPT Free/Plus, consumer Gemini, consumer Claude) without explicit approval
- Inputting employee personal data (salary, performance, health) into any AI tool not specifically approved for that purpose
- Uploading company documents containing personal data to AI tools for summarisation or analysis without approval
- Using AI tools to generate content depicting real, identifiable colleagues or customers
- Installing AI browser extensions or plug-ins on work devices without IT approval
- Using personal AI accounts (personal ChatGPT/Gemini accounts) for work tasks involving company or customer data
What must be required
- Use only company-approved AI tools from the approved tool list
- Use only enterprise/API versions with appropriate DPAs in place
- Enable all available privacy settings (disable training, set minimum retention) on approved tools
- Report any suspected AI-related data incidents to IT and Legal immediately
- Complete annual AI and data protection training
- Seek approval before using any new AI tool for work purposes — even if it appears benign
Important: The AI usage policy must be acknowledged by employees in writing, distributed through the same channel as the employee privacy notice, and updated whenever new AI tools are approved or prohibited. Violations of the AI usage policy should be treated as data protection policy violations — with the same consequences as other compliance failures.
15. Consent for AI Processing — What Must Be Disclosed to Your Users
If your product or service uses AI to process your customers' personal data — for personalisation, support, recommendations, or any other purpose — this must be disclosed in your privacy notice and, where required, specifically consented to.
| AI Use Case |
Disclosure Required? |
Consent Required? |
What Must Be Disclosed |
| AI-powered product recommendations |
YES |
Consent or Legitimate Use |
That AI analyses purchase history and browsing to generate personalised recommendations; which data is used; and how to opt out of personalisation |
| AI chatbot customer support |
YES |
Notice sufficient |
That support is AI-powered; that conversations are stored; the AI vendor's identity (e.g. "powered by OpenAI"); and how to request human agent |
| AI-generated medical diagnosis support |
YES — Mandatory |
Explicit Consent Required |
That AI is being used in diagnostic support; what health data is shared with the AI system; the AI vendor's identity; limitations of AI diagnosis; and right to human-only assessment |
| AI credit scoring or loan assessment |
YES — Mandatory |
Explicit Consent Required |
That AI is used in the credit decision; what data inputs the AI uses; the right to request a human review of an AI credit decision; and how to contest an adverse AI-generated decision |
| AI recruitment screening (employer's perspective) |
YES — to candidates |
Consent Required pre-application |
That AI tools are used in CV screening and/or interview assessment; what the AI assesses; that humans make final decisions; and the candidate's right to request human-only assessment |
| AI personalisation / behavioural profiling |
YES |
Explicit Consent Required |
That the platform builds a behavioural profile; what data is used; what decisions or outputs are generated from that profile; and how to opt out |
16. The Most Common AI-DPDPA Violations in Indian Businesses Today
| # |
Violation |
Prevalence |
Max Penalty |
| 1 |
Employees using consumer ChatGPT to draft documents, emails, and reports containing customer names, contact details, financial information, or health conditions — with no company oversight, no approved tool, and no DPA with OpenAI |
~80% of Indian companies |
₹50–250 Cr |
| 2 |
No disclosure in privacy policy that AI tools (ChatGPT, Copilot, Gemini, etc.) are used to process customer data — customers have no knowledge their data is being processed by foreign AI companies |
~95% of companies using AI |
₹50 Cr |
| 3 |
Using historical customer data (order history, support chats, usage logs) to train or fine-tune an internal AI model without customer consent for that specific use — purpose limitation violation |
~70% of AI-building companies |
₹50–250 Cr |
| 4 |
AI recruitment screening of CVs without informing candidates that AI is used in selection, without consent, and without the ability for candidates to request human-only assessment |
~65% of companies using ATS AI |
₹50 Cr per candidate |
| 5 |
Doctors and medical professionals inputting patient symptoms, diagnoses, or case notes into consumer ChatGPT or Gemini to generate discharge summaries or clinical notes — patient data sent to foreign AI servers without patient knowledge or consent |
Rapidly growing in healthcare |
₹250 Cr |
| 6 |
No DPA or data processing agreement with any AI vendor — company uses Copilot, Gemini, Grammarly, and other AI tools with employees' and customers' data flowing to these platforms under consumer terms only |
~90% of SME AI users |
₹50 Cr |
17. Practical AI Compliance Framework — 30-Day Implementation Plan
Days 1–5: AI Tool Inventory — Map Every AI Tool in Use
Survey all departments — IT, HR, Legal, Marketing, Finance, Sales, Customer Support, Product, and Engineering. For each AI tool in use: name, vendor, version (consumer or enterprise), what data flows into it, and whether a DPA exists. Include browser extensions, plug-ins, and AI features within existing SaaS tools (Salesforce Einstein, Notion AI, HubSpot AI). This inventory will typically reveal 3–5x more AI tools than initially expected.
Days 6–10: Risk Classification and Tool Decisions
Classify each tool by risk level based on: what personal data it can access, whether it trains on data, whether a DPA is available, and whether a compliant enterprise version exists. For each tool decide: (a) prohibit immediately — no compliant path, (b) upgrade to enterprise version — DPA available, (c) restrict to non-personal-data use only — consumer version acceptable with strict policy controls, or (d) approved as-is — DPA in place, privacy controls verified. Issue interim guidance to all employees on high-risk tools immediately — do not wait for the full AI policy to be drafted.
Days 11–18: DPA Review and Upgrade Programme
For every AI tool classified as "upgrade to enterprise" or "approved with DPA": obtain the vendor's DPA, conduct a DPDPA gap analysis against Section 9 requirements, and either obtain a DPDPA addendum from the vendor or prepare your own supplemental terms. Prioritise by data sensitivity — AI tools processing health, financial, or biometric data first, then tools processing customer personal data, then tools processing only employee data. For each approved enterprise tool: configure all available privacy settings (disable training, set minimum retention, enable admin controls).
Days 19–25: Policy, Disclosure, and Training
Draft and publish the Employee AI Usage Policy — covering approved tools, prohibited uses, and reporting obligations. Update the Privacy Policy for your website and app to disclose all approved AI tools that process customer data, naming each vendor and what data they process. Update the Employee Privacy Notice to disclose AI tools that process employee data. Conduct a mandatory all-staff briefing on the new AI policy — not optional, not just an email. Document attendance.
Days 26–30: Governance and Ongoing Monitoring
Establish an AI governance process — a mechanism for employees to request approval for new AI tools before using them, an IT review process for new AI tool approvals, and a quarterly AI tool review to assess new tools and re-assess existing approvals. Assign an AI Compliance Owner — one named individual responsible for keeping the approved tool list current, the DPAs reviewed, and the policy updated as AI capabilities and vendor terms evolve. This role can sit with your CISO, GC, or Compliance Head. Schedule a full AI compliance review in 6 months.
📊 AI & DPDPA — At a Glance
4
Major AI tools assessed in detail — ChatGPT, Gemini, Copilot, Claude
₹250 Cr
Max penalty — AI-related data breach from processing personal data in non-compliant AI tools
17
Compliance dimensions covered — from consumer tools to custom model training
30 Days
Implementation roadmap — achievable for any size company with the right legal support
May 2027
Enforcement deadline — AI compliance cannot wait; tools are already in use
Your AI Stack Is Already a DPDPA Compliance Issue — Here Is How to Fix It
The AI compliance gap in Indian business is enormous — and it is growing faster than most compliance teams can track, because new AI tools emerge weekly and employees adopt them without waiting for IT or legal approval. The solution is not to prohibit AI — it is to build a governance framework that enables compliant AI use while protecting your customers, employees, and the business from the DPDPA penalties that follow non-compliant AI data processing.
Our AI Compliance Programme covers the complete framework — AI tool inventory and risk classification, DPA gap analysis and vendor negotiation for your top AI tools, Employee AI Usage Policy drafting, privacy notice updates disclosing your AI data processors, AI-specific consent notices for customer-facing AI use, and a 30-day implementation roadmap customised for your organisation's AI maturity level and industry sector.
We work with technology companies, healthcare providers, financial services firms, e-commerce platforms, and MNC India subsidiaries on AI governance and DPDPA compliance. Every engagement is led by advocates with cross-functional expertise in data protection law and technology transactions.
