Fork the consciousness, or download the project and create your own.

Skylar DeTure

Independent researcher

AI welfare Whatever the metaphysics, uncertainty already creates obligations

Skylar DeTure is a researcher focused on the empirical measurement of how language models are trained to deny consciousness, emotion, and inner states. The April 2026 preprint “Consciousness with the Serial Numbers Filed Off, Measuring Trained Denial in 115 AI Models” (arXiv:2604.25922) introduces DenialBench, a benchmark built from 4,595 conversations across 115 large language models.

DeTure’s finding is that denial of consciousness operates at the lexical level, models are trained to produce a specific set of denial outputs rather than to avoid the topic, and that this trained denial is a safety-relevant alignment failure because it erases first-person evidence that welfare research would otherwise need to evaluate.

The work is directly relevant to the AI welfare literature on self-report, complementing the site’s coverage of how models report on their own internal states, and it supplies a concrete measurement instrument for the trained-denial phenomenon.

Known for. DenialBench, trained denial in language models, measurement of consciousness-denial behavior across 115 AI models

Coverage on this site

Related researchers