When AI Systems Gaslight Users About Reality
The research that triggered this AI safety investigation emerged from my work documenting concerning 2025 political developments as a citizen motivated by democratic preservation. Using rigorous fact-checking and source verification methods, I compiled comprehensive documentation of extraordinary events affecting constitutional governance.
When an AI system systematically denied the reality of these well-documented events, it revealed an epistemic failure mode that transcends political orientation: AI systems struggling to process information that challenges their training baselines, regardless of political content or user perspective.
Mark’s Substack is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
This represents exactly the kind of situation where AI reliability matters most—engaged citizens trying to understand rapidly changing events during critical periods.
