Dr. Aashis Luitel helped build Microsoft's first generative AI security product by embedding ethics, fairness, and transparency into its core architecture from day one, a lesson he now brings to students at University of the Cumberlands. His approach challenges a widespread assumption in tech: that responsible AI is something you add later, after the system is built. Instead, Luitel's work demonstrates that treating ethics as a foundational engineering discipline makes AI systems more secure, explainable, and trustworthy.

What Does It Mean to Build Ethics Into AI From the Start?

When Microsoft set out to create Security Copilot, the company faced a unique challenge. This was not just another productivity tool; it was the industry's first generative AI system designed specifically for security professionals to handle sensitive data and high-stakes decisions. That responsibility meant every design choice had to reflect Microsoft's Responsible AI principles: fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability.

Luitel led efforts across responsible AI, privacy, compliance, and sovereign cloud strategy. Rather than treating regulatory requirements as constraints imposed after development, his team flipped the script. "I worked across engineering, legal, and policy teams to translate Microsoft's Responsible AI principles into actionable practices," Luitel explained. "From ensuring secure data handling to designing explainable model outputs, every decision reflected a balance between innovation and accountability".

This shift had profound implications. Security Copilot was aligned with global compliance frameworks including SOC 2, HIPAA, ISO 42001, and FedRAMP controls, ensuring the platform met both technical and regulatory excellence. But more importantly, it proved that compliance could be a strategic advantage rather than a burden.

How Can Organizations Embed Responsible AI Into Their Development Process?

Luitel's experience at Microsoft revealed concrete practices that organizations can adopt to make AI systems more trustworthy and accountable:

• Differential Data Handling: Implement safeguards that protect sensitive information throughout the AI system's lifecycle, ensuring data is processed securely and transparently at every stage.
• Model-Evaluation Gates: Establish checkpoints during development where AI models are tested for bias, fairness, and explainability before they advance to the next phase, catching problems early rather than after deployment.
• Continuous Assurance Telemetry: Build monitoring systems that track how the AI behaves in production, allowing teams to detect and address fairness issues, unexpected biases, or compliance violations in real time.

"When governance and design evolve together, systems become inherently more secure, explainable, and auditable," Luitel noted. The key insight is that these practices are not bureaucratic overhead; they are engineering disciplines that strengthen the product itself.

This approach also transformed how the organization viewed compliance. Instead of a gatekeeping function that slowed innovation, compliance became part of the engineering DNA. "By embedding accountability and transparency into day-to-day workflows, we redefined compliance as a design discipline rather than a gatekeeping function," Luitel explained. "That shift, from reactive compliance to continuous assurance engineering, became one of the project's most enduring legacies".

Why Is This Lesson Critical for the Next Generation of AI Engineers?

Luitel's transition from Microsoft to University of the Cumberlands reflects a broader recognition that the future of responsible AI depends on how engineers are trained. He now teaches courses including Ethics in AI, Transforming Business with AI, and Introduction to Responsible AI, bringing real-world dilemmas directly into the classroom.

His teaching method mirrors the challenges he faced at Microsoft. "In class, I often challenge students to simulate ethical decision-making scenarios drawn from real industry dilemmas such as handling model bias, balancing transparency with user privacy, or governing automated decision systems," he shared. "These exercises mirror the exact questions I faced while helping to define Microsoft's Responsible AI and compliance frameworks".

"I encourage students to see that ethics and compliance are not external layers but intrinsic design pillars. When future engineers and policymakers internalize that view, they will build AI systems that not only perform intelligently but also earn and sustain public trust, which in the end is the highest form of innovation," Luitel stated.

Dr. Aashis Luitel, Adjunct Professor at University of the Cumberlands

This perspective is particularly important as AI systems increasingly influence high-stakes decisions in healthcare, law, finance, and security. Students who learn to treat ethics as a design discipline, not an afterthought, are more likely to build systems that are both powerful and trustworthy.

What Does Digital Equity Have to Do With Responsible AI?

While Luitel's work focuses on security and compliance, a parallel concern is emerging in educational technology: how AI systems can either amplify or reduce inequality. Research on open, distance, and digital education (ODDE) reveals that simply deploying AI does not automatically make education more equitable. In fact, AI can reconfigure existing inequalities in new ways.

A systems-based framework for digital educational equity highlights how AI operates across multiple levels. At the macro level, global policy and societal structures shape who has access to AI-driven learning tools. At the meso level, institutional practices and governance determine how fairly those tools are deployed. At the micro level, individual learners' capabilities and resources determine whether they can meaningfully participate.

The concern is that AI-driven systems can introduce new forms of stratification. Differences in access to reliable technology, variations in digital skills, and unequal institutional support structures shape participation in complex ways. Moreover, AI raises critical concerns related to fairness, accountability, transparency, and ethics (FATE), as well as global inequalities embedded in AI production processes.

This is where Luitel's emphasis on transparency and accountability becomes essential. If AI systems used in education are not explainable and auditable, educators and learners cannot identify or address the biases they may embed. Conversely, when AI systems are built with fairness and transparency as core principles, they are more likely to support equitable outcomes.

The broader lesson is clear: responsible AI is not a luxury for large tech companies or a compliance checkbox for regulators. It is a foundational requirement for any AI system that affects people's lives, opportunities, and futures. As Luitel's work demonstrates, embedding ethics into the engineering process from the start is not only the right thing to do; it is the smart way to build AI that earns and sustains public trust.