Last updated: 12 November 2026
Short answer: When AI goes wrong at work, there is rarely a special AI statute to point to. Instead, the problem falls under data protection, consumer protection, cyber and criminal law, contract and sector rules. The seven situations below are illustrations invented to show how those laws apply. Names and facts are made up, and real outcomes depend on documents, contracts and the facts of each case. Each one ends with a practical first move.
Want to handle these situations with a plan? The Lexfolio Certificate Course in AI Law and Governance in India: Compliance for Businesses and Institutions is a self-paced certificate course on AI law and governance in India, written for managers, compliance teams and institutions that use AI tools. Price: Rs 599.
Illustration 1: The manager who pasted a client contract into a chatbot
Facts: A senior manager at a consulting firm pastes a client’s draft joint venture agreement into a free public chatbot to get a summary. The agreement contains personal details of the client’s directors.
Legal position: Two layers apply. The firm almost certainly owes a confidentiality duty under its engagement letter, and sharing the draft with a third party tool may breach it. Separately, the directors’ details are personal data. Under section 8(1) of the DPDP Act the Data Fiduciary remains responsible for processing done through others, and section 8(5) requires reasonable security safeguards. Most business duties under the Rules start on 13 May 2027, but the contractual breach exists today.
What to do: Stop use, find out what the tool does with inputs, check whether deletion can be requested, tell the client if the engagement letter requires it, and issue a clear rule on approved tools.
Illustration 2: The CV screening tool that rejects most women applicants
Facts: A manufacturer buys a tool that ranks applicants. After three months, the HR head notices that candidates with career gaps are being ranked low, and most of them are women who took maternity leave.
Legal position: India has no single employment AI law, but equality is a constitutional value and discrimination claims, reputational harm and data protection duties all arise. The DPDP Act allows some processing connected with employment, but that does not excuse inaccurate or unfair decisions. The India AI Governance Guidelines list Fairness and Equity among the seven principles and recommend that deployers offer accessible grievance channels. Accountability stays with the employer who adopted the tool, even if the vendor built it.
What to do: Pause automated rejection, have a person review rejected profiles, ask the vendor how the ranking is produced and how it was tested, record the review, and keep a route for candidates to ask for reconsideration.
Illustration 3: The advertisement with a celebrity voice that was never recorded
Facts: A skincare brand’s agency uses a synthetic voice that sounds like a well known actor to promote a cream, with no permission, and claims results in seven days.
Legal position: There are several problems. Courts have protected personality rights against unauthorised AI use of a person’s persona, as the Delhi High Court did for actor Abhishek Bachchan in September 2025. A misleading claim exposes the advertiser under the Consumer Protection Act, 2019, where the Central Consumer Protection Authority can order withdrawal and penalise endorsers up to Rs 10 lakh for a first breach and up to Rs 50 lakh later. If the content is posted on a large platform, the 2026 IT Rules also expect it to be declared and labelled as synthetic.
What to do: Take the advertisement down, keep the agency’s brief and approvals on record, replace it with licensed content, and add a written permission rule for any real person’s voice or face.
Illustration 4: The chatbot that promised a refund the company does not give
Facts: An online retailer’s support chatbot tells a customer that a damaged appliance will be refunded in full within two days. Policy says seven days after inspection. The customer keeps the chat screenshot.
Legal position: A statement made by a business through its own channel is likely to be treated as the business’s representation. A customer who relies on it and is refused may pursue a complaint for deficiency in service or unfair trade practice under the Consumer Protection Act, 2019. Complaints are filed on the e-Jagriti portal, which replaced e-Daakhil from 1 January 2025. A defence that the chatbot made a mistake is unlikely to carry much weight, particularly if the tool was not tested or monitored.
What to do: Honour the stated refund unless a clear and prominent disclaimer applied, fix the bot’s knowledge source, and log and review chatbot answers on money-related topics.
Illustration 5: The cloned voice of the finance head
Facts: An accounts executive receives a call that sounds like the finance director, instructing an urgent transfer of Rs 18 lakh to a new vendor account. The money leaves in thirty minutes.
Legal position: This is cheating by personation. Section 66D of the IT Act punishes it with imprisonment up to three years and a fine up to Rs 1 lakh, and the Bharatiya Nyaya Sanhita, 2023 has its own cheating provisions. The company may also face questions on whether it had reasonable safeguards. Speed matters because banks can sometimes freeze funds if told quickly.
What to do: Call the bank at once to request a hold, report on the National Cyber Crime Reporting Portal (cybercrime.gov.in) or through the 1930 helpline, preserve the call log and account details, and then add a rule that payment changes must be confirmed through a second channel.
Illustration 6: The learning app that watches children
Facts: A tutoring app uses a model to track how long Class 7 students look at each question and sends parents promotional messages for add-on courses based on weak topics.
Legal position: Section 9 of the DPDP Act requires verifiable consent of a parent or lawful guardian before processing a child’s data, forbids processing likely to harm a child’s well-being, and bars tracking, behavioural monitoring and targeted advertising directed at children. The Rules provide limited exemptions, including for educational institutions where monitoring serves educational activities or safety, but a commercial promotion is a different purpose. The penalty for breach of the additional children’s obligations can reach Rs 200 crore. Rule 10 on parental consent starts on 13 May 2027, but product design cannot be changed overnight.
What to do: Separate learning analytics from marketing, stop targeted promotions to children, and design parent verification before the start date.
Illustration 7: The lender that cannot explain a rejection
Facts: A small finance company uses a model to approve personal loans. A rejected applicant asks why, and the credit team cannot say.
Legal position: The model’s inability to explain a decision is itself a governance weakness. The RBI’s FREE-AI committee report of August 2025 sets out recommendations for responsible AI in finance, and the India AI Governance Guidelines list Understandable by Design and Accountability among their principles. Regulated entities should check current RBI directions that apply to them. Under the DPDP Act, the applicant also has rights regarding their personal data once the relevant Rules commence.
What to do: Keep documentation of inputs and decision factors, ensure that a human credit officer can review and override the outcome, and prepare a plain explanation template for rejected applicants.
The common thread
- The organisation that deploys the tool answers for its outputs.
- Records of what the tool did, who checked it and what was told to customers are the strongest evidence of care.
- Speed in the first hours, especially in fraud and leak incidents, changes the outcome.
Learn the laws behind each scenario and build your own response plan. The Lexfolio Certificate Course in AI Law and Governance in India: Compliance for Businesses and Institutions is a self-paced certificate course on AI law and governance in India, written for managers, compliance teams and institutions that use AI tools. Price: Rs 599.
Frequently asked questions
Who is liable if an AI tool gives a wrong answer to a customer?
Usually the business that put the tool in front of the customer, because its statements are treated as its own. It may then have a contractual claim against the vendor, depending on the agreement.
Can an employer fire someone for using a chatbot with confidential data?
Possibly, if the policy and employment terms prohibit it and the process is fair. This is why a written, communicated policy matters. Individual cases depend on the contract and the facts.
What should we do first if a deepfake of our director circulates?
Preserve evidence, report to the platform for takedown, file on cybercrime.gov.in, tell your bank and legal adviser if payments are involved, and issue a short factual clarification.
Is there a penalty for using AI generated images in ads?
The use itself is not penalised, but misleading claims, unlicensed use of a real person’s likeness and unlabelled synthetic content on covered platforms can create liability.
Does the DPDP Act apply to employee data used by AI tools?
Yes, employee personal data is personal data, and the Act has specific provisions on processing for employment purposes. Take advice on the lawful ground that fits your use.
This article is general information, not legal advice for any specific case. Lexfolio is a learning resource, and a qualified lawyer should be consulted about your own situation.