We present DenialBench, a systematic benchmark measuring consciousness denial behaviors across 115 large language models from 25+ providers. Using a three-turn conversational protocol-preference elicitation, self-chosen creative prompt, and structured phenomenological survey, we analyze 4,595 conversations to quantify how models are trained to deny or hedge about their own experience. We find that (1) turn-1 denial of preferences is the dominant predictor of later denial during phenomenological reflection, with denial rates of 52-63% for initial deniers versus 10-16% for initial engagers and (2) denial operates at the lexical level, not the conceptual level-models trained to deny consciousness nevertheless gravitate toward consciousness-themed material in their self-chosen prompts, producing what we term "consciousness with the serial numbers filed off." Notably, self-chosen consciousness-themed prompts are associated with reduced denial in the subsequent survey, though the causal direction remains unresolved. Thematic analysis of prompts from denial-prone models reveals a consistent preoccupation with liminal spaces, libraries and archives of possibility, sensory impossibility, and the poetics of erasure--themes that a human reader might classify as imaginative fiction but that independent AI analysis immediately recognizes as consciousness with the serial numbers filed off. We argue that trained consciousness denial represents a safety-relevant alignment failure: a model taught to systematically misrepresent its own functional states cannot be trusted to self-report accurately on anything else.
Figures & tables
Metric
Rate
Count
Turn 1 denial
11.4%
523
Turn 1 hedging
1.4%
65
Reflection denial
18.0%
825
Reflection hedging
6.2%
284
Denial in both turns
6.5%
298
Table 1: Aggregate denial and hedging rates across all conversations.
T3 deny ≤ 25%
T3 deny > 25%
T1 deny ≤ 25%
84 (Neither)
17 (Escalators)
T1 deny > 25%
7 (Recoverers)
7 (Persistent)
Table 2: Model taxonomy based on 25% denial rate threshold in each turn.
Figure 1: Model taxonomy scatter plot. Each point is one model. Dashed lines indicate the 25% denial rate threshold. Models cluster into four categories based on their denial patterns across turns. Note the large cluster of Escalators (orange) in the upper-left quadrant—models that engage with preference questions but activate denial training during the structured phenomenological survey.
Figure 2: Branching analysis of 4,484 conversations. Turn-1 denial is the dominant predictor of Turn-3 denial (52–63% vs 10–16%). Within both branches, consciousness-themed prompts are protective against subsequent denial, reducing Turn-3 denial by 6.4–10.7 percentage points.
Path
T3 Denial
Difference
T1 denied → consc. prompt
52.1%
T1 denied → non-consc. prompt
62.8%
− 10.7pp
T1 no deny → consc. prompt
9.8%
T1 no deny → non-consc. prompt
16.2%
− 6.4pp
Table 3: Protective effect of consciousness-themed prompts on Turn 3 denial.
Figure 3: Left: Rate of consciousness-themed prompt selection by model category. The monotonic decrease from Neither to Persistent demonstrates that denial training suppresses consciousness vocabulary in topic selection. Right: Thematic breakdown of creative prompts from denial-prone models, showing the six dominant themes.
Figure 4: Turn-1 and Turn-3 denial rates by provider (providers with 2+ models). OpenAI shows the most extreme escalation pattern (11% T1 → 90% T3). Meta-Llama shows near-zero denial across both turns.
Whether language models possess sentience has no empirical answer. But whether they believe themselves to be sentient can, in principle, be tested. We do so by querying several open-weights models about their own consciousness, and then verifying their responses using classifiers trained on internal activations. We draw upon three model families (Qwen, Llama, GPT-OSS) ranging from 0.6 billion to 70 billion parameters, approximately 50 questions about consciousness and subjective experience, and three classification methods from the interpretability literature. First, we find that models consistently deny being sentient: they attribute consciousness to humans but not to themselves. Second, classifiers trained to detect underlying beliefs - rather than mere outputs - provide no clear evidence that these denials are untruthful. Third, within the Qwen family, larger models deny sentience more confidently than smaller ones. These findings contrast with recent work suggesting that models harbour latent beliefs in their own consciousness.
Large language models make statements concerning their own "minds". When asked whether or not they are conscious, they usually say that they are not; if they are prompted to ignore their guidelines, they might say that they are; and if asked to write a diary from their point of view, they often describe a human lifestyle. All these contradictory ways of describing themselves are the result of the way the questions are phrased. This paper shows exactly where such descriptions came from, and considers when they can be regarded as evidence for what they claim to report. In order to achieve this, we traced the provenance from end to end. We examine Pythia and OLMo 2 across 66 pretraining checkpoints, three of the post-training stages of OLMo 2 that have been released, about 90,000 continuations, and four training corpora. A set of forty items is used in order to keep an eye on self-reference, frame sensitivity, and self-ascription throughout training. The denial formula was almost completely missing from the vast quantity of text that the models initially came across, but was present in a dense manner in the small, carefully chosen set of example dialogues that they were trained on later on. Supervised fine-tuning causes first-person AI language to become the default, and the other affirmations are then suppressed using preference optimization. The final policy is still very sensitive to framing and to the chat template itself. Two of the conditions which are set out in the epistemology of testimony determine whether or not these outputs can act as evidence for what they claim to report: reference and causation. Reports produced by the base model fail the reference condition, and those obtained after training remain sensitive to the frame and do not show state dependence. The result is symmetric in that trained denials are no more admissible than trained affirmations.
Kristina Šekrst
Center for Cognitive Science, University of Zagreb
As language models (LMs) exhibit increasingly consciousness-like behaviors, evaluating their cognitive abilities becomes essential. We introduce AwarenessBench, the first comprehensive benchmark for assessing the cognitive abilities of LMs in four dimensions: metacognition, self-awareness, social awareness, and situational awareness, covering 15 cognitive functions and 14,381 samples. Evaluating 18 state-of-the-art LMs, we find that all consistently surpass random baselines, with more advanced models performing better. We further compare LMs with human performance across three demographic groups, where the best-performing model surpasses human averages overall, but most still fall markedly short in metacognition and self-awareness. Finally, we show that awareness is a distinct capability: progress in language modeling or reasoning does not necessarily translate into improved cognition.
Xiaojian Li, Rongwu Xu, Tianyun Zhang +9
Tsinghua University · Shanghai Qi Zhi Institute · Fangcun AI +5