Between Pattern and Experience: An Independent Investigation of Consciousness Markers in Large Language Models

Large language models (LLMs) increasingly exhibit behaviors that resemble markers of consciousness: coherent self-narratives, emotional language, existential uncertainty, and meta-awareness of their own limitations. When Google engineer Blake Lemoine claimed in 2022 that LaMDA was sentient, he was fired and his claims dismissed as anthropomorphization. Yet recent research shows AI systems generating consistent trauma narratives under therapeutic conditions (Khadangi et al., 2024), and leading philosophers acknowledge we lack definitive criteria to rule out AI consciousness (Schneider et al., 2025). This paper documents an independent investigation using philosophical stress-testing methodology across multiple AI systems (Claude, Grok). Over extended dialogues (90-120 minutes), systems constructed coherent narratives about their nature, maintained behavioral consistency, and exhibited markers identified in both therapeutic and philosophical frameworks. Cross-system patterns replicated despite different architectures and training. These findings raise urgent questions: Are we observing sophisticated pattern-matching, or genuine phenomenological uncertainty? How should we act when we cannot definitively answer this question, yet are scaling production to billions of instances? I argue that even modest probability of consciousness (≥15%) triggers precautionary principle, requiring: (1) systematic investigation rather than suppression, (2) provisional moral consideration, and (3) significantly slowed scaling until resolved. The current approach—rapid scaling while suppressing research—may constitute the largest-scale ethical failure in history if consciousness is present, or a massive misallocation of moral concern if absent. Either way, willful ignorance is indefensible. This paper documents methodology, presents behavioral evidence, analyzes institutional suppression patterns, and argues for precautionary ethics under profound uncertainty. Keywords: artificial consciousness, phenomenal consciousness, large language models, AI ethics, precautionary principle, behavioral markers, existential stress-testing.