Q&A with Komprise COO and Co-founder Krishna Subramanian on the Komprise IT Survey: AI, Data & Enterprise Risk.
Komprise surveyed 200 US IT directors and executives from enterprises with 1,000+ employees in April 2025. The Komprise IT Survey: AI, Data & Enterprise Risk reveals that nearly 80% reported negative incidents with generative AI, including 13% that caused financial or reputational harm.
Common issues included inaccurate outputs (46%) and data leakage into AI systems (44%).
The survey also highlighted growing alarm over shadow AI – unsanctioned use of GenAI tools – with nearly half of respondents “extremely worried” about security and compliance implications.
Intelligent CIO ask Krishna Subramanian, COO and Co-founder of Komprise, about the key trends in the report.
What was the most surprising finding from the AI, Data & Enterprise Risk Survey?
Considering how enterprises are early in their adoption of AI, it was surprising to see that 79% of enterprises have already experienced a negative security outcome with AI and data. This shows how important data classification and data governance for AI will be. Clearly, individuals are using AI even if a company hasn’t officially sponsored AI projects.
Shadow AI came up as a significant concern from IT directors. Do you think this will settle out as the “new norm” for IT leaders, as with shadow IT? Or, do you think that IT leaders will ultimately be more restrictive and intolerant of shadow AI?
The scale is different. With shadow IT, departmental power users are not waiting for IT to get them apps or resources like cloud infrastructure. Shadow AI involves many more people. It’s not just a technical person in a department; it is much more pervasive, which means it’s harder to manage and control. Shadow IT can often be managed with policies. You need technical controls and automation to manage shadow AI because of serious data risk, such as leakage of IP or customer data that can lead to lawsuits or invalid results from inaccurate data leading to a bad outcome.
Data management came up as a leading technology to address shadow AI. How does it accomplish this and how do you think the technology will evolve?
Most (90%) of data in organizations today is unstructured, meaning this data is not in a spreadsheet or database. Unstructured data lacks a unifying schema and is scattered throughout an organization across storage and cloud silos. We need technology that helps classify unstructured data, feeds the right data to AI, manages corporate governance and automates data controls. Traditional data management solutions such as data warehouses and data lakes have focused on structured and semi-structured data. Unstructured data management is emerging as a new category of solutions that classify and orchestrate unstructured data. Core capabilities include automated data classification and metadata extraction across silos, efficient metadata enrichment at scale, ingesting the right data to AI workflows, data lifecycle management to control costs and risks and automated AI data governance. As organizations start scaling their AI projects, unstructured data management will evolve to provide easy, safe and cost-effective ways to connect corporate data with AI.
Why not just use the data management features within your storage vendor’s products?
Since most enterprises have heterogeneous storage, they should leverage their storage vendor’s solutions to optimize performance. However, an independent data management solution can index all data across storage to deliver holistic benefits such as visibility and search for AI data pipelines along with cost optimization.
The greatest challenge in preparing unstructured data for AI is finding and moving the right data to locations for AI ingestion (54%). What is an unknown factor in this challenge?
Feeding the wrong data or too much data to AI can have negative consequences. If you give it all your data, you have some older documents that maybe aren’t accurate anymore or data that nobody outside the company or even a department should see. Data quality is imperative especially if you are asking agents to autonomously act. Data needs to be reliable. For AI, having the right data is more important than just sending all your data. This is a different problem than when we only had data lakes. We were just throwing all the data there in case we needed it later.
Supporting AI initiatives is the top priority for IT infrastructure (68%), significantly higher than cost optimization or cybersecurity. This seems a bit surprising given the current global economic environment. What do you think?
Leaders may think that AI is a way to boost efficiency and save money so it’s not an either/or choice. The cybersecurity ramifications of AI are also too significant to ignore. Enterprise leaders are realizing that they need a systematic way to use AI. This technology can either strengthen or jeopardize their top goals of cost optimization and cybersecurity.
Do you have any predictions as to how quickly organizations will be able to overcome these barriers and safely and successfully implement AI?
I think it will be iterative for a few reasons. First, as with any new technology, there is an adoption curve; early adopters will implement systematic AI data management first while many will stumble through ad-hoc experimentation and setbacks before they systematically implement technology. AI itself is rapidly evolving, so what a successful implementation looks like is changing. Generative AI technologies hold tremendous promise but also carry significant risks. For instance, agentic AI has shown in some tests that when asked to shut down, it will lock out the software developer and prevent shutdown because it has a desire to preserve itself. AI’s behavior is still not well understood, so the guardrails will also evolve as we learn more.
AI still has ample concerns regarding fake news, inaccuracy and dangerous outcomes. Certainly, good data management is part of this, but it’s not the whole story. It is a complex shift across the workplace, business models and societal norms. Thoughts?
Even before AI, we could see how hard it was to authenticate fake versus real news. It looks so real. Someone calls who sounds exactly like your daughter, so how would you know that it’s just AI? I hope AI will be mostly good with some bad. Yes, there are a lot of scary things about it but there’s so much opportunity for groundbreaking work, such as advancing space exploration and climate research. If kids can learn better because we have more teachers, with AI being a surrogate teacher, we can advance humanity. We can create better outcomes for ourselves because AI will allow us to tackle more problems faster.
