Indian workers embrace AI, but shadow AI puts firms' IP at security risk

Employees are using personal AI accounts alongside enterprise tools, creating security and compliance gaps, said cybersecurity experts

Artificial intelligence (AI) is becoming part of everyday work in Indian companies, but employees are increasingly using personal AI accounts and applications alongside enterprise-approved platforms. The practice, commonly referred to as “shadow AI”, is creating new security and compliance risks for organisations.

 

The concern is not simply which AI tools employees are using, but what they are putting into them. Source code, regulated data and intellectual property are among the types of information increasingly being exposed through AI applications, according to the Netskope Threat Labs Report: India 2026.

 

The report found that 82 per cent of workers in India directly interact with AI applications, while 97 per cent use software-as-a-service (SaaS) applications with AI features. It also found that 92 per cent use applications that leverage user data to train models.

 

Cyber expert Amit Jaju said most Indian companies were prioritising AI adoption, but only a small proportion had formal AI governance or ethics frameworks in place.

 

“This means employees are experimenting with AI tools faster than security and compliance teams can build guardrails. This lag is the single biggest reason shadow AI has spread so quickly across Indian workplaces,” said Jaju.

 

Tarun Wig, co-founder and chief executive officer of Innefu Labs, said restricting AI use was not the answer because most employees using unauthorised tools were looking to complete their work faster rather than deliberately bypass security controls.

 

He said companies needed visibility into the AI tools employees were using and the data being shared with them, along with real-time controls to prevent sensitive information from leaving corporate systems.

 

“The goal isn't to restrict AI, it's to make the safe path the obvious one,” Wig said.

 

Source code is the biggest exposure

 

For Indian organisations, source code has emerged as the most exposed category of sensitive information in AI-related data policy violations. Netskope found that source code accounted for 49 per cent of such violations in India over the past year.

 

Regulated data and intellectual property each accounted for 23 per cent, while passwords and application programming interface (API) keys made up another 5 per cent. Indian organisations recorded more than 3,000 AI-related data policy violations a month on average.

 

Jaju said shadow AI created an immediate data security and compliance risk because sensitive information shared with public AI tools could move outside an organisation’s control.

 

“In our forensic investigations we have already seen cases where proprietary code and confidential business data surfaced outside the organisation's control after being run through unsanctioned AI tools, and for regulated sectors like BFSI and healthcare this can trigger direct compliance violations under India's DPDP Act,” he said.

 

Developers are among the employees who may have to share sensitive information with AI tools as part of their work. The report said they commonly use AI for debugging, explaining code and writing functions, which can involve sharing source code, error messages or configuration details.

 

For an employee, pasting such information into an AI application can be a productivity shortcut. For an organisation, however, it can expose proprietary code, business logic or credentials beyond the company’s control.

 

The risk is not limited to standalone AI chatbots. AI features integrated into everyday software can also process user data, making such activity harder for security teams to identify and monitor. 

 

Personal AI accounts create a visibility gap

 

One of the biggest challenges for IT teams is distinguishing approved AI use from personal use. Employees may turn to personal accounts because they are familiar with them or face fewer restrictions, but this can determine whether prompts, files and other information are covered by corporate security policies.

 

Netskope’s India 2026 report found that personal AI use at work fell from 79 per cent to 41 per cent over the past year. During the same period, the use of organisation-managed AI tools increased from 30 per cent to 77 per cent.

 

However, 18 per cent of users now switch between personal and enterprise accounts, showing that companies still face difficulty maintaining visibility across AI usage.

 

Jaju said employees often chose personal AI accounts because of speed and convenience rather than an intention to cause harm.

 

Approved enterprise AI tools can take longer to provision, impose greater restrictions or may not support the particular task an employee wants to complete. That could range from debugging code and summarising a contract to preparing a client presentation.

 

“When there is a gap between what employees need and what IT has sanctioned, people default to whatever free consumer tool gets the job done fastest and that gap is exactly where shadow AI thrives,” he said.

 

Companies move from blocking AI to monitoring it

 

As AI use spreads across workplaces, companies are increasingly looking at ways to identify which applications employees are using, monitor the information being shared and prevent sensitive data from leaving corporate systems.

 

Data loss prevention (DLP) is becoming an important part of this approach. Netskope said the use of AI-specific DLP controls among Indian organisations increased from 42 per cent to 52 per cent over the past year.

 

The approach does not necessarily require companies to prevent employees from using AI. Instead, organisations can allow approved tools while identifying activities that pose a risk, such as uploading source code, customer information or intellectual property to an unauthorised application.

 

Both Jaju and Wig said blocking AI outright was not the answer. Companies need to identify the tools employees are using, understand what data is being shared, provide effective approved alternatives and use monitoring and DLP controls to prevent sensitive information from leaving the organisation.

 

AI agents could expand the risk

 

The shadow AI problem could become more significant as companies move from chatbots towards AI agents.

 

Traditional AI applications generally respond to a user’s prompt. AI agents can perform a wider range of tasks, including interacting with enterprise systems, accessing information and taking actions on behalf of users.

 

“Human-in-the-loop approval should remain mandatory for any agent action that touches financial systems, regulated data or irreversible business decisions,” Jaju said.

 

The risk increases when AI systems are given permission to modify information or interact with critical enterprise systems.

 

The Netskope report found that 53 per cent of organisations give AI tools write access to cloud productivity and collaboration platforms. Another 40 per cent give them access to email, while 25 per cent give them access to code repositories.

 

This makes identity and access controls increasingly important. An employee pasting source code into a chatbot can create a data leakage risk. An AI agent with permission to access a code repository or modify enterprise systems could create a wider operational risk.

 

Indian companies face an AI governance gap

 

For Indian enterprises, the challenge is no longer whether employees will use AI, but whether companies can make that use visible and manageable.

 

Organisations need to identify approved and unauthorised AI applications, distinguish personal accounts from enterprise accounts and monitor the sensitive information being shared. As AI tools become embedded in everyday workflows, security policies also need to be practical enough for employees to follow without creating barriers to productivity.

 

“Policy has to come first, but only if it is specific and enforced, not a generic circular nobody reads,” Wig said.

 

He said monitoring without policy amounted to surveillance without purpose, while policy without monitoring amounted to paperwork.

 

“I would sequence it as policy, then DLP, then monitoring layered on top for accountability. For regulated sectors like BFSI, defence and government, this sequence is not optional. A single prompt with the wrong data pasted into an unauthorised tool can mean a breach that is very hard to undo,” he said.

 

Jaju said more mature Indian enterprises were moving away from outright blocking towards structured visibility.

 

“The more mature Indian enterprises are shifting from outright blocking to structured visibility, running discovery exercises to map every AI tool and agent in use, classifying them by data sensitivity, and offering fast, secure enterprise alternatives so employees are not forced to choose between productivity and compliance,” he said.

 

He added that companies were combining this approach with AI-focused monitoring and DLP controls, as well as amnesty policies that encourage employees to disclose the AI tools they are already using without fear of penalties.

 

For Indian companies, the challenge is therefore moving from controlling access to AI towards controlling how AI is used. As employees use AI across coding, documentation, communication and other daily tasks, the distinction between approved and shadow AI will increasingly depend on whether organisations can provide secure tools that are as accessible and useful as the consumer applications employees are already familiar with.