Vijay Raina stands at the forefront of enterprise SaaS technology, bringing a wealth of expertise in software architecture and thought leadership to the cybersecurity domain. As a specialist in managing the intricate layers of modern software ecosystems, he has observed a profound transformation in how companies interact with digital tools. Today, he shares his perspective on the rising tide of Shadow AI—a phenomenon that is not just about unapproved apps but represents a systemic identity crisis for the modern enterprise. We examine the transition from human-centric IT to a world where machine actors hold the keys to the kingdom, exploring the critical governance gaps and the evolving regulatory landscape across the APAC region.
How has the fundamental nature of shadow IT changed from the unauthorized file-sharing tools of the past to the agentic AI workflows we see today?
For over a decade, we operated under a predictable playbook where shadow IT meant an employee downloading a rogue file-sharing app or spinning up a marketing dashboard without telling the IT department. Security teams could manage this through relatively static measures like web gateways, firewalls, and simple application blacklists. However, the rise of shadow AI has completely dismantled that framework because we are no longer dealing with isolated pieces of software; we are dealing with integrated, agentic workflows. When a team connects an AI assistant to their enterprise calendar or a CRM to automate their daily tasks, they aren’t just using a tool; they are handing over persistent credentials and elevated data access to machine actors. This shift is palpable because these integrations create “living” connections that operate autonomously, often bypassing the traditional perimeter defenses we’ve relied on for years.
Can you elaborate on the systemic crisis of Non-Human Identities and why they represent a greater risk than the data exposure we usually focus on?
The conversation around AI risk has been far too narrow, focusing almost exclusively on data leaks while ignoring the silent proliferation of Non-Human Identities, or NHIs. Every time an employee creates an account for a new AI tool, they aren’t just creating a login; they are establishing a new identity with its own specific access scope and session history. These tools generate API keys, tokens, and secrets that often end up buried in configuration files or browser extensions, completely invisible to the security team. These service accounts frequently exist outside the standard lifecycle management process, meaning they don’t benefit from routine credential rotation or access reviews. Because an AI agent with access to a production database carries the same privileged risk as a human administrator, these unmanaged NHIs become persistent, high-speed entry points for attackers that remain active long after the human user has moved on from the company.
What do the latest statistics regarding data breaches tell us about the actual financial and operational impact of shadow AI in 2026?
The financial consequences are no longer theoretical; they are massive and growing at an alarming rate. According to the 2026 Cost of a Data Breach Report, 43 percent of breached organizations have experienced an incident directly linked to shadow AI, which is a staggering jump from just 20 percent a year ago. These aren’t just minor inconveniences, as the average cost for a breach involving these unmanaged AI tools has climbed to a significant US$5.39 million. Beyond the dollar figures, there is a deep operational trauma that occurs when an organization realizes its own automated scripts have been working against its security interests. It is a sobering reality that nearly half of all breaches now have an AI component, highlighting that our speed of adoption is currently outstripping our ability to defend the identities we are creating.
Looking at recent incidents across the APAC region, how are these unreviewed AI integrations manifesting as real-world security failures?
We have seen several high-profile cases recently that illustrate just how quickly things can go wrong when AI tools are used without oversight. In Japan, Panasonic Holdings had to disclose a breach where an employee inadvertently uploaded the personal records of 334,000 individuals to an unsanctioned generative AI platform. Similarly, in Singapore, the food retailer Bee Cheng Hiang saw the exposure of over 95,000 customer email addresses because of an unreviewed AI-generated bulk mail script that went rogue. Even in the public sector, a contractor for the NSW Reconstruction Authority in Australia uploaded a spreadsheet to ChatGPT that exposed the sensitive health and personal records of 3,000 flood recovery applicants. These incidents aren’t just technical glitches; they are failures of identity governance where machine speed was used to process sensitive data without the necessary human guardrails.
Why is there such a significant governance gap in the APAC region compared to global averages, and how does the lack of Privileged Access Management play into this?
The governance gap in APAC is quite stark, with 53 percent of cybersecurity leaders in the region identifying AI-related NHI management as their most significant weakness. This is 10 points higher than the global average, suggesting that while the region is adopting AI at a blistering pace, the security infrastructure is struggling to keep up. A major part of the problem is that only 38 percent of APAC organizations have fully deployed a Privileged Access Management solution, leaving the vast majority of companies without a way to monitor or control high-stakes machine identities. Furthermore, 47 percent of respondents admit they simply don’t have visibility into the AI tools their employees are using. This lack of visibility, combined with the absence of robust PAM tools, creates a perfect storm where unauthorized machine actors can operate with total anonymity.
How are regional policymakers in places like Singapore and Japan shifting their stance to address the specific challenges of agentic AI?
Policymakers across the region are moving away from general data advice and toward specific frameworks that treat AI agents as entities requiring their own identities and limits. In Japan, the Ministry of Economy, Trade and Industry recently updated its AI Guidelines for Business to specifically define and address AI agents as distinct operational risks. Singapore’s IMDA has been particularly proactive, launching a Model AI Governance Framework for Agentic AI earlier this year that insists every agent must have a traceable identity linked to a responsible human. These guidelines recommend a “minimum necessary” approach, where agents are strictly limited in the tools and data they can access. While many of these frameworks, including the National AI Centre’s guidance in Australia, remain voluntary for now, they are setting the gold standard that boards and auditors are already using to measure corporate responsibility.
What practical steps should a security team take today to move toward a “zero trust” environment for both human and machine identities?
The first step is a move away from the “block and forget” mentality, because blocking a tool just forces employees to find a stealthier, unmanaged workaround. Instead, security teams need to implement continuous discovery to find every sanctioned and unsanctioned AI tool and the NHIs they’ve created. We should treat every AI agent and service account with the same discipline as a human hire, enforcing automated credential rotation and deprovisioning that triggers the moment a project ends or an employee departs. Ideally, organizations should move toward a “zero standing privilege” model, where an AI agent is only granted the elevated access it needs for a specific task and that access is automatically revoked the second the task is complete. By recording and monitoring these privileged sessions in real time, you turn a hidden risk into a governed, visible identity that follows the core principles of zero trust: verify every request and grant only the bare minimum privilege required.
What is your forecast for the future of identity management in an AI-driven enterprise?
The future of identity management will be defined by the total convergence of human and machine governance into a single, unified fabric. By the end of 2026 and heading into 2027, we will see the “identity perimeter” become the only perimeter that matters, as traditional network boundaries continue to dissolve under the weight of decentralized SaaS and autonomous agents. We are moving toward a reality where “identity” is no longer a static attribute but a dynamic, behavior-based state that is constantly re-evaluated at machine speed. Organizations that successfully transition to this identity-first model will be able to harness the incredible productivity of AI agents without the fear of losing control over their most sensitive data. Ultimately, the winners in this landscape will be those who stop treating AI as a separate security category and start treating it as the most powerful, and therefore most privileged, user in their entire environment.
