The Copilot, Not the Autopilot: Manjeet Rege’s Pragmatic Take on Ethical AI

Artificial intelligence is no longer just the quiet engine humming behind your Netflix recommendations or managing narrow algorithms behind the scenes. Over the last three years, generative AI and large language models have pushed the technology dead into the mainstream, transforming how we write, create music, and process data. Yet, as these systems scale at a breakneck pace, the conversation is shifting from what AI can do to how we manage what it shouldn't do. Dr. Manjeet Rege, a seasoned data scientist and director of the Center for Applied Artificial Intelligence at the University of St. Thomas, recently stressed that this explosive integration requires a parallel surge in consumer awareness and intentional guardrails. According to an interview featured by the University of St. Thomas, Rege points out that as AI establishes a permanent foothold in our workplaces and personal lives, users must actively build AI literacy and understand exactly where their data goes.

This push for ethical oversight isn't just academic theory for Rege, who co-authored a definitive text on the matter, Ethics and Governance of Artificial Intelligence: Frameworks, Risks, and Society, which hit shelves in early 2026. His core philosophy treats AI as a powerful copilot rather than an autopilot, an approach that becomes vital as automated tools touch sensitive sectors like workforce management and personal finance. It is an essential distinction; while AI excels at parsing huge datasets, budgeting, and running hypothetical what-if scenarios, it lacks a human moral compass and genuine situational awareness. Letting these systems run completely unsupervised introduces massive liabilities, from baked-in algorithmic bias to serious security loopholes. When an enterprise or an individual hands over the keys to an unmonitored algorithm, they aren't just automating a workflow—they are outsourcing accountability.

The High Stakes of Data Memory and Governance

A major flashpoint in the ethical AI debate centers on data privacy and the long memory of modern machine learning models. Rege warns that uploading sensitive information, such as proprietary corporate data or personal financial statements, to public cloud-hosted AI tools is a massive gamble. Once information enters these sprawling models, pulling it back out is nearly impossible. This permanent data trail creates a structural vulnerability where private user information can be inadvertently exposed or leveraged in ways consumers never intended. It's why robust corporate governance and standardized regulatory guardrails are becoming central to tech operationalization rather than being treated as an optional, late-stage afterthought.

Balancing Regional Innovation and Global Standards

The solution isn't to choke out innovation, but to ground it in structured, transparent data architecture. Rege’s recent field research and international collaborations highlight a growing global desire for practical, efficient AI frameworks that protect workforce stability while simultaneously maximizing technical utility. As states navigate this transition, the regional focus is shifting toward establishing localized AI centers that optimize data demands sustainably. Ultimately, building an ethical framework means creating systems that respect human values, mandate strict transparency, and ensure that human developers remain firmly in control of the off switch.

Behind the Scenes: The Invisible Friction in the Race for Compliant AI

The push for ethical AI is hitting a massive roadblock: the friction between engineering teams racing to ship products and compliance officers trying to manage risk. In the tech industry, speed is everything, and taking time to audit data training sets for historical bias or intellectual property violations slows down product launches. Tech giants and startups alike frequently cut corners during data collection, treating compliance as an afterthought rather than a core engineering requirement. This structural rush creates a corporate culture where safety teams are often isolated from the primary software development lifecycle.

This internal divide is further complicated by a profound global regulatory fragmentation that leaves multinational corporations in a state of limbo. While the European Union enforces strict, sweeping legislation through its AI Act, the United States relies on a patchwork of state-level laws and sector-specific federal guidelines. This regulatory mismatch forces compliance officers to constantly rewrite data governance policies to match the stricter jurisdictions. Engineering teams find themselves building entirely separate pipelines for different regions, which drastically inflates development costs and limits the scalability of new software deployments.

The historical roots of this problem stem from early machine learning models, which were trained on massive, uncurated scrapes of the public internet. These early datasets absorbed systemic biases regarding race, gender, and socioeconomic status, which are now deeply embedded in legacy algorithms. Modern developers are realizing that scrubbing these foundational models is nearly impossible without completely retraining them from scratch, a process that costs millions of dollars. As a result, businesses are forced to deploy superficial patches, attempting to filter out biased outputs on the front end rather than fixing the systemic flaws in the data core.

Enterprise stakeholders are also grappling with the economic reality of data sovereignty and cloud dependency. Relying on public, cloud-hosted AI vendors exposes corporations to vendor lock-in and unpredictable API pricing structures. To mitigate this vulnerability, forward-thinking enterprises are investing heavily in smaller, proprietary on-premise models that operate entirely within local networks. This shift allows corporations to maintain absolute control over their data footprint, eliminating the risk of proprietary code or customer information leaking into public training pools.

Ultimately, the transition toward truly ethical artificial intelligence requires a total overhaul of modern software procurement policies. Companies can no longer simply trust the marketing claims of AI vendors regarding data cleanliness and safety metrics. Independent, third-party algorithmic auditing is rapidly emerging as a mandatory industry standard for risk management. Only through rigorous, external verification can organizations protect themselves from regulatory fines, reputational damage, and the invisible liabilities hidden within unverified neural networks.

Reading Between the Lines: The Illusion of Voluntary AI Guardrails

The tech industry's sudden enthusiasm for ethical AI frameworks deserves a healthy dose of skepticism. For years, Silicon Valley operated under the mantra of moving fast and breaking things, treating regulatory oversight as an existential threat to innovation. Now, corporate boardrooms are suddenly flooded with self-imposed ethical charters, internal oversight committees, and public pledges for responsible tech deployment. This pivot looks less like a genuine moral awakening and more like a tactical maneuver designed to preempt strict government regulation. By establishing weak, voluntary guidelines, companies can project social responsibility while continuing to deploy high-risk models without real accountability.

A glaring contradiction lies in how these companies handle intellectual property and data rights. Tech executives frequently deliver polished keynote speeches about fairness and transparency, yet their legal teams are simultaneously fighting tooth and nail in court to defend the unauthorized scraping of copyrighted material. The corporate argument relies on a double standard: AI models must be allowed free access to public human creations to learn, but the resulting commercial tools must be aggressively gatekept behind expensive proprietary licenses. This dynamic concentrates data wealth into fewer hands, directly contradicting the egalitarian rhetoric used to market these systems to the public.

Furthermore, the current corporate obsession with narrow algorithmic bias often serves as a convenient distraction from broader societal harms. While engineers spend thousands of hours tweaking text outputs to ensure political correctness, they quietly ignore the massive ecological footprint of the data centers powering those very same computations. The immense energy and water consumption required to train and run large-scale language models represents a tangible, immediate climate tax. This environmental degradation is happening right now, far away from the sterile, virtual debates about future artificial general intelligence scenarios.

Looking ahead, the long-term implication of this unchecked deployment is a steady degradation of institutional trust. As deepfakes become virtually indistinguishable from reality and automated content farms flood the internet with algorithmic noise, finding verified information will become increasingly difficult. Organizations that rush to replace human customer service agents, writers, and analysts with cheap automated alternatives are trading long-term brand equity for short-term margin improvements. They risk alienating a consumer base that is growing increasingly fatigued by synthetic interactions and algorithmic dead ends.

"We are told that artificial intelligence will soon solve humanity's greatest existential crises, yet we currently use its peak capabilities to generate mediocre marketing copy, automate bad customer service, and hallucinate legal briefs. If AI is truly the fire of the twenty-first century, we seem remarkably intent on using it primarily to roast marshmallows while the house burns down around us."

Artūras Malašauskas is an AI Systems Integrator with 20+ years of production-grade web engineering experience. He has designed, shipped, and scaled enterprise Python/PHP systems for logistics, SaaS, and public-sector clients. For the past year, he has focused exclusively on AI integrations: deploying open-source LLMs, building generative media pipelines (image, audio, video), and engineering multi-agent workflows for real production environments. His standard: reproducibility, security, cost-efficient inference—no vaporware. He documents and evaluates emerging AI tooling, separating verified capabilities from marketing noise. Technical editor at: muza-ai.eu, ai-verslas.lt, ai-naujinos.lt Connect on LinkedIn