{"id":1476,"date":"2026-11-16T07:27:35","date_gmt":"2026-11-16T07:27:35","guid":{"rendered":"https:\/\/clt.nliu.ac.in\/?p=1476"},"modified":"2026-11-16T07:27:35","modified_gmt":"2026-11-16T07:27:35","slug":"invisible-processing-visible-liability-shadow-ai-and-the-accountability-gap-under-indias-digital-personal-data-protection-framework","status":"publish","type":"post","link":"https:\/\/clt.nliu.ac.in\/?p=1476","title":{"rendered":"Invisible Processing, Visible Liability: Shadow AI and the Accountability Gap under India\u2019s Digital Personal Data Protection Framework"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\">Khushi Bhati is student at Dr Ram Manohar Lohiya National Law University<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Introduction<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">It\u2019s a Saturday evening, 15 May 2027, and an employee is still at work trying to finish his task before the deadline. As usual, he uploads a batch of consumer complaints containing their personal data into a public artificial intelligence (\u201cAI\u201d) chatbot to categorise and summarise them. There is no such malicious intention on his part; he was simply trying to meet his deadlines and improve his productivity. This has become an ordinary workplace situation with as many as <a href=\"https:\/\/news.microsoft.com\/en-in\/92-of-indian-knowledge-workers-use-ai-in-the-workplace-finds-microsoft-and-linkedin-2024-work-trend-index\/\">92% of knowledge workers<\/a> in India using AI in their daily work, against 75% globally. However, what is different this time is that the <a href=\"https:\/\/www.meity.gov.in\/static\/uploads\/2025\/11\/53450e6e5dc0bfa85ebd78686cadad39.pdf\">Digital Personal Data Protection Rules, 2025<\/a> (\u201cthe Rules\u201d) are now fully in force for 2 days, and his workplace shortcut might have exposed his employer to a liability of up to <a href=\"https:\/\/www.meity.gov.in\/static\/uploads\/2024\/06\/2bf1f0e9f04e6fb4f8fef35e82c42aa5.pdf\">\u20b9250 crore<\/a> under the <a href=\"https:\/\/www.meity.gov.in\/static\/uploads\/2024\/06\/2bf1f0e9f04e6fb4f8fef35e82c42aa5.pdf\">Digital Personal Data Protection Act, 2023<\/a> (\u201cthe Act\u201d). In addition to this penalty, a breach carries its own cost, which, as the <a href=\"https:\/\/in.newsroom.ibm.com\/India-Records-its-Highest-Average-Cost-of-a-Data-Breach-2026\">IBM report<\/a> puts it, is \u20b925.5 crore for Indian organisations in 2026.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This use of AI tools by an employee without their organisation\u2019s approval or control is generally known as <a href=\"https:\/\/www.tandfonline.com\/doi\/full\/10.1080\/07366981.2026.2679653\">Shadow AI<\/a>. It creates two overlapping risks, one related to confidentiality and intellectual property, and the other to personal data breach, and it is the second one that this article addresses. Under the Act, responsibility for protecting personal data rests with Data Fiduciaries, the entities that determine the purpose and means of processing, but what happens when an employee makes the decision to use AI tools without the employer\u2019s knowledge or authorisation? This creates an accountability gap between the increasingly decentralised manner of AI adoption in the workplace and the centralised allocation of legal responsibility, as under the Act, organisations may remain liable for data processing even when it is outside their oversight and control.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Shadow AI: The Invisible Workplace Risk<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">For years, organisations have struggled with the use of unauthorised technological shortcuts, such as the use of unauthorised hardware, software, or cloud-based services by employees at the workplace, commonly known as <a href=\"https:\/\/journals.sagepub.com\/doi\/10.1177\/0266382118760143\">Shadow IT<\/a>. As AI has become widely prevalent, employees have also started using it as an everyday workplace tool to draft, analyse, or summarise documents and perform other routine tasks. However, when such use of AI tools occurs without the organisation\u2019s authorisation or knowledge, it is known as Shadow AI. While Shadow AI is said to have evolved from Shadow IT, its <a href=\"https:\/\/onlinelibrary.wiley.com\/doi\/10.1002\/jsc.2682\">\u201cgenerative, opaque and autonomous nature\u201d<\/a> introduces new risks related to data privacy, given the capability of these AI tools to process personal and confidential data. As a result, the risk has increased: from the mere use of unauthorised IT applications within an organisation\u2019s controlled environment to leaking personal data and confidential information outside the organisation\u2019s control.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This growing reliance on Shadow AI is driven by the changing realities of workplace technology adoption and the increasing pressure on employees to manage their workloads, <a href=\"https:\/\/link.springer.com\/article\/10.1007\/s42979-025-03962-x\">meet tight deadlines and compete<\/a> under the rising performance expectations in an AI-enabled workplace. But the organisation\u2019s own provision for technological adoption often fails to keep up with the employees\u2019 rising demand for such tools and the pace at which work changes, leaving them to rely on public AI tools to meet their needs. Today, around <a href=\"https:\/\/www.microsoft.com\/en-us\/worklab\/work-trend-index\/ai-at-work-is-here-now-comes-the-hard-part\">78% of AI users bring their own AI tools<\/a> into the workplace, often despite knowing that their organisation prohibit or discourages it, which means employees, not the organisation, are now deciding which AI tools enter it. This disconnect between employee-led AI adoption and organisational governance policies results in a <a href=\"https:\/\/onlinelibrary.wiley.com\/doi\/10.1002\/jsc.2682\">\u201cgovernance drift zone\u201d<\/a>, where formal policies continue to exist but fail to keep pace with the informal, evolving nature of AI adoption.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This governance gap was visible in 2023, when Samsung\u2019s engineers <a href=\"https:\/\/www.darkreading.com\/vulnerabilities-threats\/samsung-engineers-sensitive-data-chatgpt-warnings-ai-use-workplace\">uploaded confidential source code and internal meeting<\/a> notes into ChatGPT. Following this incident, the company <a href=\"https:\/\/www.forbes.com\/sites\/siladityaray\/2023\/05\/02\/samsung-bans-chatgpt-and-other-chatbots-for-employees-after-sensitive-code-leak\/\">restricted the use of generative AI tools<\/a> due to concerns that information submitted to the platform was retained by OpenAI for model training and could be exposed to third parties. What would happen when the information leaving the organisation\u2019s controlled environment includes a consumer\u2019s personal data instead of source code?<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>The DPDP Act\u2019s Hidden Premise<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Organisations cannot govern what they cannot see, yet the Act is based on an <a href=\"https:\/\/www.barandbench.com\/columns\/the-dpdp-act-section-85-blindspot-use-of-ai-by-employees-and-the-unseen-ip-data-breach\">implicit assumption that Data Fiduciaries have sufficient visibility<\/a> over the processing of personal data. <a href=\"https:\/\/indiankanoon.org\/doc\/141683147\/\">Section 8(5)<\/a> of the Act requires Data Fiduciaries to protect personal data \u201cin their possession or under their control\u201d, including processing undertaken by a Data Processor on their behalf. <a href=\"https:\/\/www.meity.gov.in\/static\/uploads\/2025\/11\/53450e6e5dc0bfa85ebd78686cadad39.pdf\">Rule 6<\/a> of the rules requires Data Fiduciaries to implement reasonable security safeguards like monitoring the use of personal data, controlling access to the systems on which it is held, and reviewing logs to detect unauthorised access. But it becomes difficult to fulfil the requirements of these provisions in scenarios like Shadow AI, where organisations have zero visibility into what, how, when and by whom the personal data is uploaded to a public AI tool. This processing can take place even when the organisation takes all reasonable security safeguards, as the employee\u2019s interaction with a public AI tool happens outside the organisation\u2019s controlled environment.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The difficulty, therefore, is that there is no express regulation that regulates the use of AI by employees at the individual level, and the DPDP framework also does not explicitly address data processing through public AI applications. Yet it imposes a responsibility on Data Fiduciaries to protect personal data. This raises a question: can an organisation avoid its responsibility under Section 8(5) when an employee independently uploads personal data to public AI tools without authorisation? <a href=\"https:\/\/indiankanoon.org\/doc\/34143931\/\">Section 8(1)<\/a> provides that a Data Fiduciary is responsible for processing undertaken by it or on its behalf by a Data Processor, \u201cirrespective of any agreement to the contrary\u201d. The stronger reading is therefore that an organisation cannot escape its liability under Section 8(5), because the employee was acting in the course of his employment and the organisation held the data until he uploaded it. Organisations may implement internal policies to restrict employees\u2019 conduct, but a policy can only be enforced where its breach can be detected, and it makes no difference whether the employee acted carelessly or on purpose, since Section 8(1) doesn\u2019t ask what he intended. This results in an accountability gap: employees control how AI tools are used, but the organisation remains responsible for that use under the DPDP framework.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This raises concerns about the AI provider\u2019s responsibility for data processing, since <a href=\"https:\/\/indiankanoon.org\/doc\/53541514\/\">Section 8(2)<\/a> allows the Data Fiduciary to engage a Data Processor \u201conly under a valid contract\u201d. That assumes the organisation chooses the Data Processor, but AI adoption is now more often led by employees than by the organisation. Where an employee is the one uploading the data to an AI application without the organisation\u2019s authorisation, and there is no contractual connection between the Data Fiduciary and the provider, responsibility cannot be cleanly shared. But such an upload might lead to a personal data breach under <a href=\"https:\/\/www.meity.gov.in\/static\/uploads\/2024\/06\/2bf1f0e9f04e6fb4f8fef35e82c42aa5.pdf\">Section 2(u)<\/a> of the Act, and <a href=\"https:\/\/indiankanoon.org\/doc\/101309009\/\">Section 8(6)<\/a> read with <a href=\"https:\/\/www.meity.gov.in\/static\/uploads\/2025\/11\/53450e6e5dc0bfa85ebd78686cadad39.pdf\">Rule 7<\/a> then requires the Data Fiduciary to notify the Data Principal without delay and the Data Protection Board of India (\u201cthe Board\u201d) within seventy-two hours of becoming aware of the breach. Where the upload was never detected, that period never begins, and the exposure lies not in a failure to notify but in the failure of the safeguards that should have made the organisation aware.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Closing the Gap: Governance That Follows How AI Is Actually Used<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The novice solution to the problem seems to be simply to ban the use of AI tools, as was done by Samsung in 2023. However, as reported by <a href=\"https:\/\/www.upguard.com\/resources\/the-state-of-shadow-ai\">UpGuard<\/a>, around 45% of the workers who came across any blocked AI tool in their organisation found other ways to continue using it. Instead, it may give employees a reason to use AI in the shadows, which will eventually make it more difficult for organisations to know about, control, or even be aware of any such use, pushing it far outside the organisation\u2019s visibility and control. Therefore, organisations need to adopt a balanced approach, such as providing employees with approved AI tools, which will allow them to retain the <a href=\"https:\/\/gprjournals.org\/journals\/index.php\/ajt\/article\/view\/537\">productivity gains<\/a> while ensuring the safety of personal data or confidential information.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This was visible in <a href=\"https:\/\/www.netskope.com\/resources\/cloud-and-threat-reports\/cloud-and-threat-report-2026\">Netskope\u2019s 2026 report<\/a>, according to which the use of organisation-managed AI accounts increased from 25% to 62% within a year, and personal account use fell from 78% to 47%.&nbsp; Approved AI use means that the purpose of its use is decided by the organisation, access runs through the organisation\u2019s own login rather than a personal account, and the tool is engaged either through a platform the organisation has built itself, or through an enterprise platform provided by a third party where the provider acts as Data processor, under a valid contract, as required by Section 8(2). These practices give the organisation the ability to monitor, log, and review the tool\u2019s use, and place contractual obligations on the provider where it acts as a Data Processor. But merely providing authorised AI tools may not achieve the intended purpose, as they may simply return to public AI tools if the authorised ones are harder to use or less capable.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This can begin with categorising data on the basis of its sensitivity, as an employee who knows the actual consequences of processing personal data will be far more careful than one who is simply asked to use an approved AI tool. This also requires separating personal data, as mentioned under Section 2(t) of the Act, from confidential information, as the two may overlap but are not the same, as one can lead to consequences to the employee for leaking or breaching it, but on the other hand, the DPDP Act does not yet provide for any such liability against the employee for data processing. Complying with Section 8(5) and Rule 6 requires organisations to adopt reasonable security measures to prevent a personal data breach, but relying on continuous monitoring to enforce these measures may, in some situations, amount to employee surveillance. While <a href=\"https:\/\/indiankanoon.org\/doc\/62814281\/\">Section 7(i)<\/a> permits processing for the purpose of employment and for safeguarding the employer from loss or liability, the extent to which measures for preventing breaches of personal data fall within it remains unclear, and monitoring beyond this permissible limit has been argued to be <a href=\"https:\/\/www.livelaw.in\/articles\/employment-surveillance-clauses-void-dpdp-regime-539957\">void<\/a> under <a href=\"https:\/\/indiankanoon.org\/doc\/1625889\/\">Section 23 of the Contract Act<\/a>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This leaves the organisation in a difficult situation where it has to create a balance between having enough visibility to detect the use of AI, but not breaching the privacy of employees by going beyond the limit of Section 7. Requiring organisations to make employees sign an agreement for the non-use of unauthorised AI tools, similar to existing NDAs for confidential information, can reduce the need for continuous monitoring. However, such an agreement does not help in transferring the Data Fiduciary\u2019s statutory responsibility under the Act, including its duty under <a href=\"https:\/\/indiankanoon.org\/doc\/60458195\/\">Section 33(2)(e)<\/a> to show timely mitigation once a breach is detected. The Act was framed without taking into consideration the increasing use of AI, but it should now clarify whether and in what scenarios the responsibility can be shared or attributed to other parties involved, instead of resting it entirely on the organisations.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Conclusion<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Shadow AI is not simply a technology problem, but a governance issue, as the Act places responsibility on the Data Fiduciary to protect personal data, yet it does not clearly address the situation where the data breach is a result of unauthorised use of AI by employees. This leaves an accountability gap between the one actually processing the personal data and the one bearing its statutory liability of up to \u20b9250 crore under the Act. The challenge, therefore, is to close that gap, which stops being theoretical by <a href=\"https:\/\/www.meity.gov.in\/static\/uploads\/2025\/11\/53450e6e5dc0bfa85ebd78686cadad39.pdf\">13 May 2027<\/a>, and it will take more than internal policies and security measures to do so. The Act needs to catch up with how AI is actually used in the workplace and make clear where responsibility lies when Shadow AI results in a breach.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Khushi Bhati is student at Dr Ram Manohar Lohiya National Law University Introduction It\u2019s a Saturday evening, 15 May 2027, and an [&hellip;]<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-1476","post","type-post","status-publish","format-standard","hentry","category-blog-series","col-md-6 col-sm-6"],"_links":{"self":[{"href":"https:\/\/clt.nliu.ac.in\/index.php?rest_route=\/wp\/v2\/posts\/1476","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/clt.nliu.ac.in\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/clt.nliu.ac.in\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/clt.nliu.ac.in\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/clt.nliu.ac.in\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=1476"}],"version-history":[{"count":1,"href":"https:\/\/clt.nliu.ac.in\/index.php?rest_route=\/wp\/v2\/posts\/1476\/revisions"}],"predecessor-version":[{"id":1477,"href":"https:\/\/clt.nliu.ac.in\/index.php?rest_route=\/wp\/v2\/posts\/1476\/revisions\/1477"}],"wp:attachment":[{"href":"https:\/\/clt.nliu.ac.in\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=1476"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/clt.nliu.ac.in\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=1476"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/clt.nliu.ac.in\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=1476"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}