Generative AI and curated Large Language Models (LLM)

As Generative Artificial Intelligence (GenAI) continues to reshape industries and government operations, organizations face an important challenge: how can they harness the power of Large Language Models (LLMs) while ensuring accuracy, security, transparency, and trust? That question was at the center of a thought-provoking panel hosted by Shivaji Sengupta, CEO of NXTKey Corporation, during AFCEA TechNet Cyber 2024 in Baltimore, Maryland.

The session, “Generative AI and Curated Large Language Models (LLMs),” brought together experts from the Defense Information Systems Agency (DISA) and the Department of Homeland Security (DHS) to explore the rapid evolution of Generative AI, its growing impact across government and industry, and the importance of developing AI systems that are not only powerful but also secure, explainable, and mission-focused.

The discussion highlighted that Generative AI has moved well beyond being an emerging technology. It is becoming a strategic capability that is transforming how organizations automate work, improve decision-making, enhance cybersecurity, accelerate software development, and deliver better services. However, with this opportunity comes the responsibility to ensure AI systems operate using trusted data, strong governance, and appropriate human oversight.

One of the first topics explored was the remarkable evolution of Generative AI and Large Language Models. In only a few years, AI models have grown dramatically in capability, enabling organizations to generate natural language, summarize complex information, write software, analyze large datasets, and assist with research at unprecedented speed. These advances have been driven by larger training datasets, increased computing power, and innovations in neural network architectures. Yet, as these models become more capable, they also become more complex, requiring careful attention to model governance, security, and operational reliability.

The panel also examined the expanding range of real-world applications for Generative AI across both government and commercial organizations. Agencies are increasingly evaluating AI to automate customer support, improve workforce productivity, accelerate document analysis, enhance knowledge management, strengthen cybersecurity operations, and support predictive analytics. Rather than replacing human expertise, the panel emphasized that Generative AI should augment human decision-making by eliminating repetitive work and allowing professionals to focus on higher-value activities that require judgment and experience.

A significant portion of the discussion focused on the ethical responsibilities associated with deploying Generative AI. As organizations adopt increasingly capable AI systems, they must also address concerns surrounding privacy, bias, misinformation, intellectual property, transparency, and accountability. Panelists discussed the risks associated with hallucinations, deepfakes, disinformation campaigns, and unintended bias within AI models. These challenges reinforce the need for comprehensive AI governance frameworks that ensure AI systems operate responsibly while maintaining public trust and regulatory compliance.

One of the most compelling themes of the session was the concept of curated Large Language Models. Unlike general-purpose LLMs trained on massive volumes of publicly available information, curated LLMs are designed using carefully selected, validated, and governed data sources that are specific to an organization’s mission and operational environment. This approach significantly improves the quality, relevance, and reliability of AI-generated outputs while reducing the likelihood of inaccurate responses, hallucinations, or exposure of sensitive information.

For government agencies and regulated industries, curated LLMs offer several important advantages. They enable organizations to apply AI to highly specialized domains while maintaining greater control over data provenance, model behavior, security, and compliance. Human oversight remains an essential component of this approach, ensuring that AI-generated recommendations are reviewed, validated, and continuously refined as organizational knowledge evolves.

The discussion also addressed the importance of balancing innovation with security. As Generative AI becomes integrated into mission-critical environments, organizations must establish strong controls around data protection, identity management, model security, access controls, and continuous monitoring. AI governance should encompass not only technical safeguards but also policies, risk management processes, workforce training, and executive oversight to ensure responsible adoption across the enterprise.

Looking toward the future, the panel explored how Generative AI is expected to become increasingly integrated into enterprise operations. Advances in multimodal AI, domain-specific language models, autonomous AI agents, explainable AI (XAI), and retrieval-augmented generation (RAG) are expected to make AI systems more accurate, transparent, and useful across a wide range of government and commercial applications. At the same time, evolving regulatory frameworks and industry standards will continue shaping how organizations deploy AI responsibly while protecting privacy, security, and public trust.

The discussion concluded with an engaging question-and-answer session that reflected the growing interest among government agencies and industry leaders in understanding both the opportunities and challenges associated with Generative AI. Audience questions centered on practical implementation strategies, governance, cybersecurity implications, and how organizations can move from experimentation to production deployments while minimizing operational and ethical risks.

The overarching message from the session was clear: the future of Generative AI will not be defined solely by increasingly powerful models, but by the quality of the data, governance, security, and human expertise that guide them. Organizations that invest in curated Large Language Models, trusted data, and responsible AI governance will be better positioned to leverage AI safely while delivering meaningful mission outcomes.

As governments and enterprises continue accelerating their AI initiatives, success will depend not only on adopting advanced technologies but on implementing them thoughtfully. Curated LLMs represent an important step toward building AI systems that are more accurate, explainable, secure, and aligned with organizational objectives. The conversations at AFCEA TechNet Cyber 2024 reinforced that responsible innovation is the key to unlocking the full potential of Generative AI while maintaining the trust and confidence of users, stakeholders, and the public.