Ongoing Study Reveals Epistemological Flaws in Gemini and Grok as Risk Factors for AI Safety and Alignment

WOODSTOCK, Vt., Sept. 16, 2026 /PRNewswire/ -- Artificial Epistemics, LLC (AE) announced today its latest findings from an ongoing study of prevailing epistemologies (knowledge production functions) in leading large language models (LLMs) – this time involving the specifics of Gemini and Grok. Using a distinction philosophers make between justificationist and falsificationist epistemologies, the AE team found that both LLMs (Gemini and Grok) explicitly rely on justificationist thinking when formulating responses to prompts or plans of action. In so doing, they systematically overstate the correctness of both facts and values and thereby undermine AI safety and alignment by exposing users to the risks and consequences of false or illegitimate information when used. In a joint statement by Joseph M. Firestone and Mark W. McElroy, co-founders of AE and creators of the Susty Code, the study's leaders had this to say about their findings:

Chatbots in AI are no different than humans insofar as the truth and morality of what they say and do is concerned. Both are irreducibly fallible. This raises the question of how AIs test and evaluate their claims before saying or doing anything. Justificationist approaches, for their part, give priority to finding supporting evidence; falsificationist methods, by contrast, look for refutations. The difference is that whereas no amount of supporting evidence can justify a claim as either true or legitimate, it only takes a single contradiction to refute one.

Our findings today are that two of the leading chatbots, Gemini and Grok, openly rely on the same notoriously fallacious form of justificationist epistemology, whereby the knowledge they produce and/or the actions they take are treated as essentially true or legitimate with certainty, thanks mainly to the authority vested in them by their makers. This more or less ignores the possibility of conflicting/contradictory evidence, and relies instead on the managerial authority of those who own or control AIs to systematically disregard the exceptions. As long as enough supporting evidence can be found, they reason, why bother looking for contradictions?

Indeed, if what we want in AI are models prone to spreading falsehoods, hallucinations, or rogue behaviors, justificationist epistemologies are perfect for the job. If, on the other hand, what we want are AIs that routinely flag and kill their worst ideas before they kill us, the opposite is true. That is the kind of reform we must have in AI at this time. Why? Because the road to AI safety and alignment must be paved with falsificationism. Nothing less will do!

The study cited above is an ongoing attempt by AE to better understand the epistemologies found in the world's leading AI models, and to advocate for what they must be in order to ensure AI safety and alignment. The company's latest white paper just released on this topic can be found here: What is the Primary Epistemology of Leading LLMs?

About Artificial Epistemics

Artificial Epistemics, LLC is a U.S.-based startup founded in early 2026, whose purpose is to help guard against the dangers of misinformation and rogue behaviors by AI. Its strategy is to work with leading AI producers to help integrate its epistemic tools into their offerings. See here for an introductory white paper on the company's primary offering, a leading falsificationist protocol: The Sustainability (Susty) Code.

Contact: Mark W. McElroy
Co-Founding Principal
[email protected]

SOURCE Artificial Epistemics, LLC