Beyond the Hype: How to Set Up Your AI Project for Real Success – Source: www.proofpoint.com

Today, AI is the golden ticket to productivity gains and competitive advantage. From automating repetitive tasks to unearthing insights hidden in oceans of data, AI seems to promise it all. But reality tells a different story. Many AI initiatives are struggling to gain traction, while others have launched only to crash and burn. Why? The short answer is this: Companies either don’t know the risks, or they underestimate them. 

This year, IBM published a piece and the title says it all: “Data Suggests Growth in Enterprise Adoption of AI is Due to Widespread Deployment by Early Adopters, But Barriers Keep 40% in the Exploration and Experimentation Phases.” Don’t kill the messenger – I am seeing it everywhere I go. 

Some organizations are cautious—they know AI introduces potential vulnerabilities and are hesitant to proceed without proper data security foundations.  

But a significant number plunge ahead, captivated by AI’s promise, only to pull the plug after a breach or inadvertent data leak. It’s a classic case of “you don’t know what you don’t know.” And with AI, the stakes are high. 

The Pitfall of Diving in Blindly

Imagine you’re building a house. You’re excited to get to the finished product, so you skip some foundational steps to speed things up. Maybe you bypass the structural engineering or cut corners on the framework. It looks great—until the first storm hits. That’s what happens when organizations jump into AI without securing their data foundations. 

AI, particularly when leveraging large language models (LLMs) and other generative tools, relies on vast amounts of data to function. But when this data isn’t classified, secured, and managed appropriately, the results can be catastrophic. 

In a recent survey, 70% of companies reported at least one AI-related security or privacy incident. In many cases, they hadn’t anticipated how data would flow or who might access sensitive information within the AI system. The result? Costly breaches and a hard lesson learned. 

The Hesitation – For Good Reason

For others, the risks are all too clear. Leaders know AI introduces new challenges to data security, regulatory compliance, and privacy. They may recognize that they lack a solid data framework but don’t know where to start in building one. And so, the project stalls, caught between the lure of AI’s benefits and the fear of unknown risks. 

One key issue is data sprawl—without proper classification and governance, data can be mismanaged or misplaced within AI systems.  

In fact, Gartner predicts that, by 2028, more than 50% of enterprises that have built their own large language models (LLMs) from scratch will abandon their efforts due to costs, complexity, and technical debt. 

For organizations weighing the risks of AI, these numbers are cause for caution. 

LLMs and the Trouble With Data Exposure

Let’s talk about large language models (LLMs), which are among the most powerful tools in AI. They’re highly effective at handling tasks from generating human-like text to automating processes.  

But here’s the catch: LLMs don’t inherently understand data sensitivity. They process whatever data they’re fed, meaning if sensitive information—like customer PII, proprietary code, or strategic insights—isn’t managed, it can easily end up exposed. 

Consider this: In 2023, a global semiconductor experienced a significant data leak involving large language models. Employees inadvertently disclosed confidential company information by using OpenAI’s ChatGPT to review and debug source code. 

The oversight wasn’t malicious; it stemmed from a lack of data classification and governance tailored for AI. When LLMs are deployed without guardrails, they can amplify risk rather than mitigating it. 

I did a session about this very thing at RSA in May of 2024 and since then, I have heard dozens of other examples. 

Building a Data-First, Human-Centric Approach

So, how do you prevent AI projects from stalling or failing? The answer lies in taking a data-first, human-centric approach.  

‘Data-first’ means understanding and classifying your data, tracking where it flows, and knowing exactly who or what has access to it. Meanwhile, the ‘human-centric’ element is essential because it’s people who train the machines, prepare the data, and ensure it’s ready to support AI securely and effectively.  

By putting data and people at the center, we build a strong foundation that makes AI both powerful and safe to deploy. 

Before implementing any AI initiative, organizations need to establish foundational data security and governance controls. Here’s a roadmap to get started: 

1. Classify Your Data: Identify and tag data by sensitivity levels (e.g., confidential, internal, public). Classification helps ensure that AI systems don’t inadvertently expose or mishandle sensitive information. 
2. Establish Guardrails: Set usage policies and permissions to limit how data flows within AI systems. Establish checks to prevent sensitive data from being used inappropriately. 
3. Build in Monitoring and Visibility: Use tools that provide visibility into how and where data moves within AI ecosystems. Monitoring can help spot unauthorized use or anomalies in real-time. 
4. Continuous Training: Educate employees on the risks associated with AI tools — particularly around handling sensitive data. Many incidents happen not due to malice but due to lack of awareness. 

Moving Forward With Eyes Wide Open

The promise of AI is enormous, but so is the risk when projects proceed without foundational security. The reality is, AI introduces both new capabilities and new vulnerabilities—many of which organizations aren’t yet prepared for. By taking a data-first approach, classifying information, and setting strong guardrails, companies can unlock AI’s potential safely. 

In this age of digital transformation, AI projects don’t have to be cautionary tales. With the right data foundation, they can drive meaningful productivity gains and insights—without the risk of costly missteps.