Reader question

An employee uploaded customer records into an AI tool. How do we contain the problem?

Answered by Advocate Sharan Jain··Cyber Crime & Online Fraud

Legal Shorts · 74 words

Stop further uploads and preserve the exact file, prompt, account details and logs. Check the AI service, account tier, terms and settings that applied at the time. Ask the provider through its official channel about access, retention and deletion, without assuming that deleting a conversation removes every copy or that training occurred. Assess current CERT-In, contractual and sensitive-data obligations promptly. As at 17 November 2026, core DPDP business duties are scheduled for May 2027.

Short sources checked:

WhatsApp

Stop further uploads, identify exactly what was sent and preserve the account and upload records. Then use the provider's verified support or administrative route to restrict access and request appropriate deletion or retention controls. Do not assume deleting the visible conversation removes every stored copy, and do not describe the data as used for model training until the relevant product terms and settings have been checked.

The response depends on the actual service, account tier, upload date and configuration. A consumer account, a managed business workspace and an API arrangement can have different terms. Record the arrangement that existed when the upload happened, rather than relying on a current marketing description of the tool.

What should we do before questioning the employee?

Secure the relevant business account and prevent further sharing. Preserve the file version, prompt, conversation identifier, time, user account and any available audit log. Ask the employee to keep the original material and explain the sequence. Do not direct them to delete the account before the business has preserved the evidence needed to understand the event.

Limit circulation of the incident file. Copying the customer workbook into a large management chat to ask what went wrong creates another disclosure. A restricted inventory of the fields and affected records can support the first assessment while the original is held securely.

Separate three questions
What did the employee upload? Who could access it? What does the provider retain or use under the actual service arrangement? Evidence answering one question does not automatically answer the others.

How do we establish the scope?

Compare the upload with its source file. Check hidden sheets, comments, attachments and connected files, as well as visible rows. Identify whether names or contact details were included, whether the data was genuinely anonymised and whether confidential business material was also present. Removing a name may leave an individual identifiable through other fields.

Find out whether the conversation was shared, published, connected to another application or used to generate a downloadable output. Preserve the relevant sharing settings. Do not repeatedly run the material through the tool to see what happens. Further experiments can expand the disclosure and contaminate the record.

What should we ask the AI provider?

Use the provider's official business or privacy channel. Supply enough identifiers to locate the incident without resending all the personal data. Ask about retention, access, downstream processing, deletion options, available logs and the limits of any deletion confirmation. Request preservation of material needed for investigation where appropriate, while distinguishing that request from continued routine use.

Read the terms and data-processing agreement for the actual product. Check account settings, administrator policies and any changes made after the event. A general statement that business data is not used for training may not establish that a personal account used by the employee had the same setting. Equally, an upload does not prove that training occurred. State the verified position.

Which legal duties need assessment today?

As checked on 17 November 2026, the ordinary DPDP processing and breach duties are scheduled for 13 May 2027. The consent-manager provisions that commenced in November 2026 do not bring the whole regime forward. Existing IT Act provisions, the 2011 sensitive personal data rules and contracts need their own assessment on the present facts.

Check whether the upload disclosed sensitive personal data, breached a customer confidentiality undertaking or sent information outside an agreed processing environment. The DPDP Act's future processor-contract and security duties are useful preparation points, but should not be described as already applicable solely because the incident involves AI. A customer's contractual restrictions can be enforceable independently of that future timetable.

Could this require incident reporting?

Assess CERT-In's current directions promptly. Covered entities must report specified cyber incidents within six hours of noticing them or being told about them. The reporting material includes data leaks and breaches. Whether this upload falls within a reportable category requires a factual assessment, including what was exposed and how. Do not assume that an employee's mistake is automatically excluded.

The CERT-In FAQs allow an initial report based on available information followed by additional details. Contractual notification duties and sector requirements may have different triggers. Record the discovery time, assessment, action and acknowledgements. Waiting for a provider to answer every question does not itself suspend an independent reporting clock.

  • The original uploaded file, prompt and conversation identifiers.
  • The account tier, applicable terms, settings and access logs.
  • The customer contract and any approved-tool policy in force at the time.
  • The provider's written response on access, retention and deletion.
  • The incident timeline and separate reporting or customer-notification assessments.

How should we deal with the employee and customers?

Investigate the facts before deciding whether the event was a mistake, a policy failure or deliberate misuse. Check whether the business had provided an approved tool, clear instructions and workable alternatives. Follow the applicable disciplinary process. Do not ask the employee to write an admission drafted to protect the company or to invent a technical explanation.

Where a customer must be notified, explain the data involved, confirmed facts, containment and remaining uncertainty. Avoid claiming that the matter is closed because the employee clicked delete. Protective advice should match the actual risk, such as compromised credentials or exposure of payment information, rather than a generic alarming message.

What should change after containment?

The useful control is a clear boundary around what may be uploaded, to which approved service and under whose authority. Give staff a practical way to remove unnecessary identifiers and obtain approval for uncertain cases. Review browser extensions and connected applications as well as the main chat interface.

My concern would be whether the business can reconstruct this upload and stop the next one. A policy that forbids everything but leaves staff without an approved route is difficult to apply consistently. Test the revised process using synthetic data, record the lesson and review the incident file's retention. The business data-protection guide covers the wider framework. The personal-data remedies answer and cyber-crime practice page address separate individual and incident routes.

Sources

The law this answer relies on, so you can read it yourself.

  1. 1.Digital Personal Data Protection Act, 2023 Read the source
  2. 2.G.S.R. 843(E), DPDP Act commencement Read the source
  3. 3.Sensitive Personal Data Rules, 2011, WIPO Lex official legislative reproduction Read the source
  4. 4.Information Technology Act, 2000, updated text Read the source
  5. 5.CERT-In directions under IT Act section 70B, 28 April 2022 Read the source
  6. 6.CERT-In FAQs on Cyber Security Directions, May 2022 Read the source

The short answer's sources were checked on 17 November 2026. Statutes and judgments can change, so check the current position before you act on anything here.

Nothing there yet? Send the question in and it gets answered here.

Related legal service

Dealing with this yourself rather than reading about it? Our Bangalore advocates work in this area.

Go deeper on this

This answer is the short version. These guides cover the same ground in full, with the procedure, the timelines and the leading cases.

SJ

Answered by

Advocate Sharan Jain

Advocate based in Bangalore, practising before the Karnataka High Court and District, Sessions, Consumer and Family courts. Answers public legal questions to make Indian law more accessible.

This answer is general information on Indian law as at November 17, 2026, published for public education. It is not legal advice, it does not take account of your facts, and reading it does not create an advocate-client relationship. Law changes and every case turns on its own circumstances. Please consult a qualified advocate about your own matter.

People also asked

Cyber & Online Fraud

A company keeps using my personal data. What rights do I have under the new data law?

The new data law is being brought into force in stages. As of September 2026, its core access, correction, erasure and grievance rights have not yet commenced. The notification schedules those provisions eighteen months after publication. You can still write to the company identifying the data and use you object to, and ask for its response under the rules currently applicable to it. Keep that correspondence. Do not assume every DPDP right is already enforceable today.

Cyber & Online Fraud

Someone is harassing and threatening me online. What can I actually do?

Save the full messages, account details and dates, then report the account to the platform and the conduct to the cybercrime portal or police. Threats intended to cause alarm can amount to criminal intimidation under Section 351 BNS. Not every rude message meets that test. Explain the pattern and quote the threats accurately. If you face immediate danger, seek urgent police help. Blocking the account can protect you, but keep an evidence copy first if safe.

Cyber & Online Fraud

We emailed a customer spreadsheet to the wrong person. What should we do immediately?

Revoke any shared link, try an administrative recall and contact the unintended recipient through a verified channel. Preserve the original email, workbook and access logs before removing working copies. Establish which customers and data fields were affected. As at 17 November 2026, core DPDP breach duties are scheduled for May 2027, while current CERT-In, sectoral and contractual reporting obligations need prompt assessment. A recall or deletion assurance reduces uncertainty but does not prove that every copy has disappeared.

Cyber & Online Fraud

I received a legal notice saying my company misused personal data. What should I check before replying?

Verify the sender, the data involved and the dates before replying. Preserve the original notice, collection wording, access logs and vendor records. As at 17 November 2026, most DPDP business duties are scheduled for May 2027, but existing IT Act, security and contractual obligations need separate assessment. A disclosed leak may also trigger current CERT-In reporting duties. Keep confirmed facts separate from matters still being investigated and have admissions or settlement commitments reviewed before the reply is sent.

Cyber & Online Fraud

The bank says I shared my OTP so the loss is mine. Is that actually the law?

Sharing an OTP can count as customer negligence, but the RBI rule is more specific than a blanket refusal. For an unauthorised transaction caused by that negligence, the customer bears the loss until it is reported. Losses after reporting fall on the bank. Other rules apply where the bank was at fault or a third-party breach caused the loss. Report immediately and ask the bank to identify the evidence and rule supporting its decision.

Cyber & Online Fraud

My UPI was used without my knowledge. I never shared anything. Who bears the loss?

Report the unauthorised UPI debit to your bank immediately and keep proof of the time. Under RBI rules, bank fault can mean zero customer liability. A third-party breach, with neither bank nor customer at fault, also attracts zero liability if reported within three working days of receiving the bank's communication. Customer negligence follows a different rule. Ask for the transaction and alert records. The bank bears the burden of proving customer liability under this framework.

S Jain & Attorneys · Ask Me

Still not the question you had in mind?

Search the column, or send your question in. Questions of general interest are answered here, anonymously, so the next person does not have to ask.