From Fear to Future: How Organizations Can Securely Embrace Generative AI

The promise of AI, specifically generative AI, has been at the forefront of many organizations’ strategic and even board-level discussions.

Generative AI promises to drive profound economic disruption, from automating an ever-growing list of complex tasks to driving innovation across all facets of business, and holding the promise of a more automated and robotic-led world.

Yet, despite promise and hype, we see very few organizations successfully integrating AI into their workflows, let alone enabling employees to engage with large language models in their day-to-day tasks.

Interestingly, even some of the most technically advanced organizations in traditional AI/ML functions, such as tech companies, banks, and government entities, still have policies in place that block the use of tools like ChatGPT.

While AI still has a long way to go in improving accuracy (or removing hallucinations, in tech jargon) and finding effective and repeatable use cases, one of the primary reasons businesses refuse employee access to GenAI models is due to data security concerns. This includes concerns around the sensitive data being shared to the third party solution providers, a chance that the model itself will train on the data and expose it to other users, or as most recently seen with DeepSeek, concerns around intentional data espionage from jurisdictions like China.

The Cloud Computing Parallel

In many ways, these concerns are following similar trajectories as seen with the introduction of cloud computing. Cloud computing offered the promise of enhancing an organization’s compute capabilities while eliminating the ongoing and costly capital expenditure required to stay up to speed with technology developments.

Yet, even with the promise of lower costs, including the ability to eliminate hardware, support, real estate, and security spend associated with on-premise servers, we still continue to see organizations hesitant to have their data stored by a third party.

While the data security provisions in a modern cloud architecture are second to none, it does not mean they are infallible, and many organizations still prefer to have their valuable data under direct control with on-premise servers.

The AI Data Security Challenge

AI is no different. While we can use tools like ChatGPT to help with low-risk tasks such as content writing, research, or project planning, the real value of Generative AI lies in embedding it within core systems to unlock insights and drive value from big data.

As such, the prospect of sharing highly secure information, from financial projections and product development to engineering IP, employee personal information, and client data, with a software system that has the potential to retain, train, and share the data with others, is a legitimate concern.

The truth is that many of the AI solutions on the market today still lack robust data security measures. Free versions of chatbots are the wild west of data security, and while paid versions may offer rights stating that the AI won’t be trained on your data, it’s still a far way from the requirements a CIO would demand.

Even newly emerging “enterprise” additions from foundation model providers like OpenAI and Anthropic should be carefully considered to ensure SOC 2 and other regulatory and data security measures to protect data are in place.

Addressing Data Security Concerns

These fears should not prevent organizations from exploring the use of AI. In fact, these are exactly the challenges organizations need to research and address. With a rich ecosystem of open-source models and cloud solutions like AWS’s Bedrock services, there are many ways to leverage AI while still maintaining full control of one’s data.

At SGK, this was our approach from day one. We knew that as a trusted partner to many of the world’s largest brands, retailers, and pharmaceutical providers, we would need to maintain 100% trusted and secure stewardship of their data.

So, we started to build out an AI infrastructure within our cloud environment that assures no data leakage and, by extension, blocks any ability for model providers to train on our data. Data sits within our dedicated storage environments, from the content or files uploaded to an AI solution to the vectorized RAG databases that make that data usable by AI systems.

By building our platform in such a manner, we have the benefit of unlocking lower consumption-based costs and decoupling the model from the data, enabling us to continually upgrade with new model releases.

The Path Forward

New AI developments coming in 2025 will position SGK to deliver local service in more countries through an expanded global framework, in turn backed by a larger, more diverse network of creative, production, and technical expertise.

Alongside large-scale investments in cutting-edge content and packaging automation software, SGK will be positioned to better serve clients to meet the rapidly growing consumer demand for technological solutions, including AI.

We are fortunate at SGK to have strong development and cloud teams, along with key partners who have helped us along this journey. With the right approach and understanding of the risks, there are clear paths to innovate with Generative AI while still working within corporate data security guardrails and policies.

Best of all, as AI continues to develop with the promise of increased intelligence and capabilities, we have an understood roadmap to quickly understand and integrate these solutions into our tech stack: plug and play, so to speak.

Rejecting generative AI due to legitimate but manageable risks will be a strategic mistake for many organizations. We are at a tipping point. No different than fax vs email, mail-catalogue vs e-commerce, the future is clear on who the winners will be.

AI software was used to help speed up the initial editing of this article. It was then honed, shaped and brought to life by the power of the human mind.