Safeguarding sensitive data: prompt engineering for Gen AI

Generative Artificial Intelligence (GenAI) represents a transformative advancement in technology with capabilities to autonomously generate diverse content, such as text, images, simulations, and beyond. While GenAI offers significant operational benefits it also introduces risks, particularly in mission-critical industries such as national defense and space. The emergence of GenAI is similar to the invention of the internet, electricity, spacecraft, and nuclear weapons. A major risk with GenAI is the potential for data reconstruction, where AI systems can inadvertently regenerate or infer sensitive mission data, even from anonymized or fragmented inputs. This is relevant today because we are in an AI arms race against our adversaries much like the race to the moon and development of nuclear weapons. Such vulnerabilities pose profound threats to data security, privacy, and the integrity of mission operations with consequences to national security, societal safety and stability. This dissertation investigates the role of prompt engineering as a strategic intervention to mitigate GenAI's data reconstruction risks. By systematically exploring how tailored prompting techniques can influence AI outputs, this research aims to develop a robust framework for secure GenAI deployment in sensitive environments. Grounded in systems engineering principles, the study integrates theoretical models with experimental analyses, assessing the efficacy of various prompt engineering strategies in reducing data leakage, bias, and confabulation. The research also aligns with AI governance frameworks, including the NIST AI Risk Management Framework (RMF) 600-1, addressing policy directives such as Executive Order 14110 on the safe, secure, and trustworthy development of AI. Through mixed-methods experimentation and stakeholder interviews within defense and space industries, this work identifies key vulnerabilities and proposes actionable mitigations. The findings demonstrate that prompt engineering, when applied systematically, can significantly reduce the risks of data reconstruction while enhancing AI system reliability and ethical alignment. This dissertation contributes to the broader discourse on Responsible AI (RAI), offering practical guidelines for integrating GenAI into mission-critical operations without compromising data security. This underscores the imperative of balancing GenAI's transformative potential with the societal need for robust safeguards against its inherent risks.

Name thesis:degree_name

Doctor of Philosophy (Ph.D.)

Level thesis:degree_level

Doctoral

Discipline thesis:degree_discipline

Systems Engineering

Grantor dc:publisher

Colorado State University. Libraries

Year dc:date.issued

2025

Authors dc:creator

• Giang, Jennifer, author
• Simske, Steven J., advisor
• Marzolf, Gregory, committee member
• Gallegos, Erika, committee member
• Ray, Indrajit, committee member

• mission data
• prompt engineering
• systems engineering
• mission engineering
• generative artificial intelligence
• risk mitigation

Statement dc:rights

• Copyright and other restrictions may apply. User is responsible for compliance with all applicable laws. For information about copyright law, please see https://libguides.colostate.edu/copyright.

Language dc:language.iso

eng, English

Identifier URI

https://doi.org/10.25675/3.02325

OAI identifier oai:identifier

oai:mountainscholar.org:10217/241093