AI's Moral Compass? A Startup's Plan to Code Truth into Machines

WOODSTOCK, VT – March 17, 2026 – In an era grappling with the dual promise and peril of artificial intelligence, a Vermont-based startup has announced a bold plan to build a conscience directly into the machine. Artificial Epistemics, LLC today unveiled its 'Susty Code,' a protocol it claims can quality-control the truth and morality of AI-generated knowledge before it ever reaches a user.

The announcement positions the company as a new contender in the high-stakes field of AI safety, aiming to tackle the rampant issues of misinformation, bias, and unethical content that plague current systems. Unlike many solutions that focus on filtering or flagging problematic outputs after they're created, Artificial Epistemics proposes a pre-emptive strike, intervening at the foundational level of knowledge processing.

A New Foundation for AI Knowledge

The core innovation of the 'Susty Code,' according to its creators, is its "pre-action" approach. While major AI labs like OpenAI and Google DeepMind employ extensive "red teaming" and post-generation safety filters, the Susty Code is designed to be an integral part of the AI's cognitive architecture. It aims to vet information and value judgments as the AI is processing them, long before they are formulated into a sentence, image, or decision.

To achieve this, the protocol integrates long-standing principles from diverse fields, including the empirical rigor of science, the structural coherence of logic, the ethical frameworks of value theory, and the study of knowledge itself—epistemology.

In a joint statement, company co-founders Joseph M. Firestone, PhD and Mark W. McElroy, PhD, described their approach as a fundamental upgrade to how AI operates. "The Susty Code adds a layer of intelligence to AI called knowledge processing, a natural function in humans and other living systems, but still missing from most of AI today," they stated. "This goes well beyond pattern-matching, probabilities, and simple reinforcement learning by bringing sophisticated epistemic rules carefully into play."

The company argues that this layer of self-regulation is the missing ingredient for creating truly safe and reliable AI, especially as these systems are granted more autonomy in critical domains.

The Crowded Market for AI Trust

Artificial Epistemics enters a market where the demand for AI governance has never been higher. A significant "trust gap" persists among the public and enterprise customers, fueled by high-profile incidents of AI-generated falsehoods, discriminatory outputs, and privacy violations. This has created a booming industry for AI safety, risk management, and compliance solutions.

The competitive landscape is populated by companies offering a range of services. Governance platforms like Credo AI provide dashboards for enterprises to monitor their AI models for risk and compliance. Fact-checking tools from firms like Winston AI and TruthScan specialize in detecting AI-generated text and deepfakes after the fact.

The Susty Code seeks to differentiate itself by being a preventative, rather than a reactive, solution. Instead of cleaning up messes, Artificial Epistemics aims to prevent the mess from being made in the first place. This unique value proposition could be compelling for organizations looking to build trust and mitigate reputational and legal risks from the ground up.

Hurdles of Integration and Adoption

Despite its ambitious vision, the path to widespread adoption for the Susty Code is fraught with challenges. The primary obstacle is technical and logistical. The AI ecosystem is a veritable "Tower of Babel," with a myriad of data formats, proprietary architectures, and a lack of standardization. Integrating a third-party protocol at the core knowledge-processing level of a complex, large-scale AI model is a monumental engineering task.

Furthermore, the industry's giants have already invested billions in their own internal safety and alignment teams. Persuading them to outsource a function as critical as their models' ethical and factual core to a startup will be a formidable sales challenge. These leading AI producers may view their proprietary safety frameworks as a key competitive advantage and be reluctant to cede control or introduce external dependencies into their tightly controlled systems. The cost, effort, and perceived risk of re-architecting their platforms around a new, external protocol would have to be justified by an undeniable leap in safety and performance.

Whose Truth? The Unavoidable Ethical Questions

Beyond the technical and business hurdles lies a deeper, more philosophical challenge. The very premise of a protocol that "quality-controls truth and morality" immediately begs the question: Whose truth and whose morality?

While the Susty Code's goals align perfectly with the spirit of emerging regulations like the EU AI Act, which demands robust, transparent, and ethical AI, the implementation is profoundly complex. The act of encoding "truth" and "morality" into a set of rules risks embedding the inherent biases and cultural perspectives of its creators. What is considered a moral imperative in one culture may be viewed differently in another. A fact deemed essential in one context might be considered misleading without additional nuance in another.

Critics of such top-down, rule-based ethical systems argue that they can be rigid and struggle to adapt to novel situations. The project of defining a universal, computationally legible set of ethical and epistemic principles is a challenge that has occupied philosophers for millennia and is unlikely to be solved by a single software protocol. As Artificial Epistemics seeks partners to integrate its code, it will inevitably face intense scrutiny over the governance of the protocol itself. Who decides what rules go in, how are they updated, and what recourse exists when the "moral code" itself is flawed?

The launch of the Susty Code represents a fascinating and potentially crucial step in the maturation of artificial intelligence. It crystallizes the central tension of the AI era: the immense potential to engineer better systems versus the profound difficulty of capturing the nuances of human knowledge and values in lines of code. The success or failure of this endeavor may offer telling insights into whether AI can be taught not just to calculate, but to comprehend.