Shadow AI in Banking: The Governance Challenge of Unauthorised AI
AI in the Workplace: The AI Risk Hiding in Plain Sight
Artificial Intelligence (AI) is rapidly becoming part of everyday work in banks. Today, it is not unusual to find employees using Generative AI tools such as ChatGPT, Google Gemini or Microsoft Copilot to draft customer emails, summarise lengthy Reserve Bank of India (RBI) circulars, prepare credit appraisal notes, analyse policy documents, or generate presentations. These tools are fast, easy to access and capable of significantly improving productivity.
The challenge, however, is that many of these AI applications are being used without formal approval from the organisation. Employees often turn to publicly available AI tools to complete routine tasks, unaware that they may be exposing sensitive business or customer information in the process. This growing use of unauthorised AI applications is commonly referred to as Shadow AI.
Shadow AI has evolved from the broader concept of Shadow IT, where employees use unapproved software or technology outside the organisation’s official IT environment. However, the risks associated with Shadow AI are far more significant. Unlike conventional software, AI systems can process large volumes of information, generate added content, infer insights from sensitive data, and interact with external cloud-based platforms. If used without appropriate safeguards, they may inadvertently expose confidential information, produce inaccurate outputs, or create compliance, cybersecurity, and reputational risks.
For Indian banks, these risks are no longer hypothetical. As financial institutions accelerate their digital transformation and employees increasingly adopt AI tools in day-to-day operations, managing Shadow AI has become an important governance challenge. The need is not to discourage innovation, but to ensure that AI is adopted within a secure, well-governed and responsible framework that protects customer trust, regulatory compliance, and institutional resilience.
Understanding Shadow AI: More Than a Technology Issue
Shadow AI refers to the use of Artificial Intelligence (AI) tools by employees without the knowledge, approval or oversight of the organisation’s technology, risk, or compliance functions. In many cases, employees adopt these tools with good intentions, such as improving productivity or completing routine tasks more efficiently. However, when AI is used outside approved governance frameworks, it can expose banks to significant operational, cybersecurity and regulatory risks.
In a banking environment, Shadow AI can take many forms. A credit officer may use a public AI chatbot to summarise a borrower’s financial statements, an HR executive may upload employee data to an online AI application for analysis, or a business team may use an AI-powered writing assistant to prepare customer communications. While these activities may appear harmless, they often involve sharing sensitive information with external AI platforms that operate outside the bank’s approved technology environment.
What makes Shadow AI more challenging than traditional Shadow IT is the nature of AI itself. Conventional unauthorised software primarily creates risks related to data storage or access control. AI tools, however, analyse information, generate new content and draw inferences from the data they receive. They can also produce inaccurate or fabricated responses that appear convincing, commonly referred to as AI hallucinations. More importantly, publicly available AI models are not designed around a bank’s internal policies, regulatory obligations, or customer confidentiality requirements. As a result, employees may unknowingly rely on inaccurate outputs or expose sensitive information, creating risks that extend far beyond technology into governance, compliance, and customer trust.
Why Does Shadow AI Thrive?
Shadow AI is rarely the result of deliberate policy violations. More often, it emerges when employees try to improve productivity in the absence of approved AI tools. Understanding these underlying drivers is essential for developing an effective governance framework.
(i) Productivity Gap: Banking professionals work under tight timelines while handling large volumes of information. AI tools can help draft documents, analyse data, and summarise regulations within minutes. When approved alternatives are unavailable, employees often turn to publicly accessible AI applications to improve efficiency.
(ii) Awareness Gap: Many employees do not fully appreciate that entering business or customer information into a public AI tool may expose sensitive data outside the bank’s secure environment. This highlights the need for continuous awareness and training on the responsible use of AI.
(iii) Approval Gap: The rapid pace of AI innovation often outstrips traditional technology approval processes. Lengthy evaluations and procurement cycles may encourage employees to adopt readily available AI tools before officially approved solutions become available.
(iv) Leadership Gap: Organisational culture also plays a key role. If productivity is encouraged without clear guidance on the acceptable use of AI, employees may assume that using unapproved tools is acceptable. Clear direction from senior management, supported by practical policies and approved AI platforms, is therefore critical to reducing Shadow AI.
The Hidden Risks of Shadow AI
The risks associated with Shadow AI extend well beyond technology. Unauthorised use of AI tools can expose banks to data privacy, operational, legal, reputational and governance risks. Since these risks are often interconnected, a single instance of Shadow AI may have implications across multiple areas of the organisation.
(i) Data Privacy and Confidentiality
The most immediate concern is the unauthorised sharing of sensitive information. Employees may inadvertently upload customer data, financial records, KYC documents or internal reports to public AI platforms, where the bank has limited visibility or control over how the information is stored or processed. Such practices may create compliance concerns under the Digital Personal Data Protection (DPDP) Act, 2023 and may also conflict with RBI’s expectations on data governance and third-party risk management.
(ii) Inaccurate or Misleading AI Outputs
Generative AI tools do not always produce factually correct responses. They can generate convincing but inaccurate information, commonly referred to as AI hallucinations. If AI-generated content is used without appropriate human verification, it may result in incorrect customer communication, flawed credit assessments, compliance lapses, or poor business decisions. The confidence with which AI presents inaccurate information makes this risk particularly challenging.
(iii) Regulatory and Legal Exposure
The use of unauthorised AI applications may expose banks to regulatory and legal risks, particularly where customer information, confidential business data or regulated activities are involved. As regulators continue to strengthen expectations around AI governance, institutions will be expected to demonstrate appropriate oversight, accountability, and control over AI-enabled processes, regardless of whether the tools are formally approved or informally adopted by employees.
(iv) Intellectual Property and Business Information
Shadow AI may also put an institution’s proprietary information at risk. Uploading internal policies, product strategies, pricing models, or confidential business documents to external AI platforms can inadvertently expose valuable intellectual property. Once sensitive information is shared with public AI models, its use and retention may be difficult for the organisation to control.
(v) Reputational and Conduct Risk
For banks, trust is one of the most valuable assets. A single incident involving unauthorised use of AI or accidental disclosure of customer information can significantly impact customer confidence and the institution’s reputation. There is also a conduct risk if employees rely on AI-generated recommendations or communications without adequate review, potentially resulting in unsuitable advice, regulatory scrutiny, or customer disputes.
Taken together, these risks demonstrate that Shadow AI is far more than an IT concern. It is an enterprise-wide governance challenge requiring coordinated oversight from business, technology, information security, compliance, legal and risk management functions.
The Regulatory Perspective
A common misconception is that Shadow AI exists in a regulatory grey area because there is no dedicated regulation governing its use. In reality, banks are already subject to a range of regulatory and legal requirements that apply irrespective of whether AI tools are officially approved or informally adopted by employees. Shadow AI therefore falls within the scope of existing expectations relating to technology governance, data protection, cybersecurity, and operational risk.
RBI’s Expectations on AI Governance: The Reserve Bank of India (RBI) has consistently emphasised that regulated entities remain accountable for technology-related risks. The Master Direction on Information Technology Governance, Risk, Controls and Assurance Practices (2023) requires banks to strengthen governance over technology assets, third-party risks, and information security. More recently, the Draft Guidance on Regulatory Principles for Model Risk Management (2026) reinforces the importance of governance, independent validation, human oversight and accountability for AI and machine learning models. Although the draft focuses on institutionally deployed models, its principles are equally relevant to the risks arising from unauthorised AI use.
Data Protection and Privacy Obligations: The Digital Personal Data Protection (DPDP) Act, 2023 places clear responsibilities on banks, as Data Fiduciaries, to process personal data only for authorised purposes and with appropriate safeguards. Uploading customer or business information to unauthorised AI platforms may therefore expose banks to legal, regulatory, and reputational risks.
Preparing for the Next Phase of AI Governance: The regulatory landscape is evolving rapidly. Initiatives such as the IndiaAI Mission, together with international developments such as the EU AI Act and guidance issued by financial regulators in the United Kingdom and Singapore, reflect a growing emphasis on responsible AI adoption, transparency, and governance. Although there is no regulation dedicated exclusively to Shadow AI, the responsibilities of banks are already well established through existing requirements on technology governance, data protection, and risk management.
Lessons from Global Experience
Shadow AI is not a hypothetical concern. Organisations across industries have already experienced how the unauthorised use of Generative AI can expose sensitive information, create governance challenges, and prompt stronger organisational controls. These experiences provide important lessons for banks.
Samsung’s Data Leakage Incident
One of the earliest and most widely reported examples occurred in 2023, when employees at Samsung Electronics inadvertently uploaded confidential source code and internal business information to ChatGPT while seeking coding assistance. The incident prompted Samsung to temporarily restrict the use of Generative AI tools on company devices and accelerate the development of internal AI governance policies. The key lesson was that once confidential information is entered into a public AI platform, the organisation may have little control over how that information is retained or processed.
Growing Regulatory Focus on AI Governance
Regulators across the world are responding by strengthening expectations around AI governance rather than issuing rules specifically for Shadow AI. The European Union’s AI Act, together with guidance from financial regulators in jurisdictions such as the United Kingdom and Singapore, emphasises governance, transparency, accountability, and human oversight for AI systems used in regulated sectors.
The growing emphasis on governance is also being shaped by recent AI safety evaluations. Leading AI developers, including OpenAI and Anthropic, published findings from controlled safety evaluations that explored how increasingly capable AI systems behaved when given complex objectives and access to external tools. The results highlighted the importance of robust governance, human oversight and effective control mechanisms as AI systems become more autonomous.
For banks, these developments highlight that AI governance is no longer limited to model accuracy or cybersecurity. It now extends to ensuring that AI systems operate within clearly defined boundaries, supported by appropriate controls, accountability, and continuous oversight.
Implications for Banks
These developments reinforce an important lesson for banks. The greatest risks associated with Shadow AI often arise not from malicious intent but from well-intentioned employees using powerful AI tools outside approved governance frameworks. As AI adoption accelerates, effective governance, employee awareness and secure enterprise AI platforms will be essential to balancing innovation with security, regulatory compliance, and customer trust.
A Practical Framework for Governing Shadow AI: The GUARD Approach
Managing Shadow AI requires more than simply restricting the use of AI tools. Blanket bans are difficult to enforce and may encourage employees to use unauthorised applications without visibility or oversight. A more effective approach is to provide approved AI solutions, establish clear governance and promote responsible use across the organisation.
A practical way to achieve this is through the GUARD Framework, which brings together five key elements of effective AI governance.
G – Governance: Define Accountability
Establish clear roles and responsibilities for AI governance, with oversight from senior management and participation from business, technology, information security, compliance, and risk functions. Governance should not only control AI adoption but also enable timely evaluation and approval of business use cases.
U – Understand AI Usage
Banks need visibility into how AI tools are being used across the organisation. Periodic assessments, employee disclosures, and technology monitoring can help identify approved and unauthorised AI applications, enabling a risk-based approach rather than a purely restrictive one.
A – Assess Risk
Not every AI tool presents the same level of risk. AI applications should be assessed based on factors such as the sensitivity of data processed, business criticality, degree of human oversight and the potential impact on customers, operations, and regulatory compliance. This enables proportionate governance and control measures.
R – Responsible Use Through Awareness
Policies alone are not sufficient. Employees should receive practical guidance on the safe and responsible use of AI, including what information should never be shared with external AI platforms, when approval is required and how to report concerns. Regular awareness programmes are essential to build a culture of responsible AI use.
D – Detect, Monitor and Respond
Continuous monitoring is critical to identifying unauthorised AI usage and preventing data leakage. Banks should strengthen technical controls, periodically review AI usage, and treat Shadow AI incidents as opportunities to improve governance, strengthen controls, and enhance employee awareness.
The GUARD framework encourages a balanced approach that supports innovation while maintaining appropriate governance, regulatory compliance, and customer trust. By combining effective oversight with responsible adoption, banks can realise the benefits of AI without compromising security or operational resilience.
Moving from Shadow AI to Trusted AI
The objective of AI governance should not be to discourage the use of AI, but to bring its use within a secure and well-governed environment. Restricting external AI tools alone may not address the productivity needs that encourage employees to use them in the first place. Banks therefore need to make the approved route to AI adoption both accessible and practical.
1. Provide Secure and Approved AI Platforms
Employees are less likely to rely on unauthorised tools when suitable alternatives are readily available. Banks should provide enterprise-grade AI solutions with appropriate safeguards for data protection, access control, information security, and regulatory compliance. This allows employees to benefit from AI while keeping sensitive information within approved environments.
2. Enable Controlled Experimentation
Banks should provide governed environments where employees and business teams can explore promising AI use cases without exposing customer or institutional data to unnecessary risk. Such controlled experimentation allows potential benefits and risks to be assessed before wider deployment, while encouraging innovation within defined boundaries.
3. Strengthen AI Literacy Across the Bank
Technology controls cannot address Shadow AI on their own. Employees need to understand what information can safely be shared with AI tools, where the risks arise, how AI-generated outputs should be verified and when human judgement remains essential. Regular awareness and role-based training can help employees become both responsible users of AI and an important line of defence against its misuse.
Moving from Shadow AI to trusted AI therefore requires a combination of secure technology, controlled experimentation, and informed employees. When these elements work together, banks can encourage productive use of AI without compromising customer information, regulatory obligations, or institutional trust.
When Shadow AI Becomes Autonomous
While many organisations are still addressing the risks associated with Generative AI, the technology is evolving rapidly. The next wave of Shadow AI is likely to emerge from Agentic AI – AI systems capable of planning, making decisions and executing multiple tasks autonomously to achieve a defined objective. Unlike Generative AI, which primarily responds to prompts, AI agents can interact with applications, retrieve information, use external tools, and complete end-to-end workflows with limited human intervention.
This shift significantly expands the Shadow AI risk landscape. An employee deploying an unauthorised AI agent to process customer requests, analyse financial data or automate business processes could inadvertently introduce an autonomous decision-making system into the bank’s operations without appropriate governance, validation, or oversight. The risks extend beyond data leakage to include unauthorised actions, operational disruptions, compliance failures, and increased cybersecurity exposure.
Although the use of Agentic AI in banking is still at an early stage, banks should begin preparing now. Governance frameworks developed for today’s Shadow AI should be flexible enough to address increasingly autonomous AI systems, supported by appropriate human oversight, model risk management and clearly defined operational boundaries. Building these capabilities today will help institutions respond more effectively as AI continues to evolve.
Responsible AI Begins with Responsible Governance
Artificial Intelligence is rapidly becoming part of everyday banking operations, and its adoption will only continue to grow. The challenge for banks is therefore not whether employees will use AI, but whether its use takes place within a secure, transparent, and well-governed environment. Unauthorised AI adoption may improve individual productivity, but without appropriate safeguards it can also expose institutions to data privacy, operational, regulatory, and reputational risks.
Managing Shadow AI does not require organisations to discourage innovation. Instead, it calls for a balanced approach that combines secure enterprise AI platforms, clear governance, employee awareness, and effective oversight. Banks that make responsible AI adoption easier than unauthorised AI use will be better equipped to realise the benefits of AI while protecting customer information, meeting regulatory expectations and strengthening operational resilience.
As AI capabilities continue to evolve, Shadow AI should be viewed not merely as a technology challenge but as an enterprise-wide governance priority. Institutions that establish robust governance today will be better prepared to adopt future AI innovations safely, responsibly and with greater confidence.
Authored by:

Anish Shrimali
Chief Manager, Union Bank of India,
Union Learning Academy-Digital Transformation, Mumbai

