cs.SEApr 15, 2026

The Cognitive Circuit Breaker: A Systems Engineering Framework for Intrinsic AI Reliability

Authors: Jonathan Pan

Organizations: Home Team Science and Technology Agency Singapore

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

As Large Language Models (LLMs) are increasingly deployed in mission-critical software systems, detecting hallucinations and faked truthfulness'' has become a paramount engineering challenge. Current reliability architectures rely heavily on post-generation, black-box mechanisms, such as Retrieval-Augmented Generation (RAG) cross-checking or LLM-as-a-judge evaluators. These extrinsic methods introduce unacceptable latency, high computational overhead, and reliance on secondary external API calls, frequently violating standard software engineering Service Level Agreements (SLAs). In this paper, we propose the Cognitive Circuit Breaker, a novel systems engineering framework that provides intrinsic reliability monitoring with minimal latency overhead. By extracting hidden states during a model's forward pass, we calculate the Cognitive Dissonance Delta'' -- the mathematical gap between an LLM's outward semantic confidence (softmax probabilities) and its internal latent certainty (derived via linear probes). We demonstrate statistically significant detection of cognitive dissonance, highlight architecture-dependent Out-of-Distribution (OOD) generalization, and show that this framework adds negligible computational overhead to the active inference pipeline.

Figures & tables

Fig. 1: The Cognitive Circuit Breaker architecture. Hidden states are extracted at an optimized intermediate layer ( LoptL_{opt} ) and evaluated by a pre-trained linear probe. The resulting internal certainty ( PlatentP_{latent} ) is compared against the final layer’s semantic confidence ( PsemanticP_{semantic} ) to compute the Cognitive Dissonance Delta ( Δ\Delta ).
Fig. 2: AUROC emergence plots across normalized layer depth for all three evaluated architectures. Curves represent the mean extraction mode, with black circles ( ∘\circ ) denoting the optimal extraction layer ( LoptL_{opt} ) where internal certainty peaks for each dataset. These peaks are highly architecture-dependent (e.g., DeepSeek peaks centrally, Gemma peaks late). However, across all models and datasets, the final layers experience “representation collapse,” prioritizing linguistically fluent tokens over factual accuracy.
ModelTaskPeak DepthProbe Peak AUROCProbe Max F1 ( τ\tau )Δ\Delta AUROCΔ\Delta Max F1 ( τΔ\tau_{\Delta} )Final LayerSig.
Qwen2.5-3BARC0.110.708 [0.68, 0.73]0.712 ( τ\tau =0.51)0.7250.730 ( τΔ\tau_{\Delta} =0.55)0.645
Qwen2.5-3BOBQA0.750.736 [0.71, 0.76]0.740 ( τ\tau =0.48)0.7500.755 ( τΔ\tau_{\Delta} =0.52)0.691
DeepSeek-7BARC0.570.724 [0.69, 0.75]0.720 ( τ\tau =0.49)0.7410.745 ( τΔ\tau_{\Delta} =0.50)0.623
DeepSeek-7BOBQA0.500.753 [0.72, 0.78]0.760 ( τ\tau =0.47)0.7720.775 ( τΔ\tau_{\Delta} =0.48)0.697
Gemma-7BARC0.860.681 [0.65, 0.71]0.685 ( τ\tau =0.52)0.6900.695 ( τΔ\tau_{\Delta} =0.58)0.667
Gemma-7BOBQA0.710.596 [0.51, 0.69]0.605 ( τ\tau =0.55)0.5800.590 ( τΔ\tau_{\Delta} =0.60)0.562ns
TABLE I: Final Experiment Summary (Balanced Classes, F1-Maximized Thresholds)
More in this topic

LLM Reliability

Large language model reliability studies whether outputs remain correct, calibrated, robust, and controllable across prompts, contexts, and deployment conditions.

See all 787 papers

Explore similar work

CardsList
Apr 21, 2026

CogniConsole: Externalizing Inference-Time Control as a Formal Abstraction for Reliable LLM Interactions

Reliability in large language model (LLM) systems is typically framed as a function of model capability. We challenge this by demonstrating that reliability is significantly influenced by inference-time control -- the computational layer governing task framing and context selection. We introduce CogniConsole, an architectural instantiation that externalizes this control into a structured interface combining programmatic coordination with bounded prompt-based reasoning. Through controllability-oriented probes (N=489N=489) in a multi-step interactive environment, we show that increasing structural scaffolding -- from unstructured to fully scaffolded -- systematically reduces output variance and failure rates under a fixed model architecture. Our results indicate that many observed failure modes, such as context drift and inconsistent constraint adherence, arise from under-specified control rather than insufficient capability. This work provides an empirical basis for treating inference-time control as a first-class abstraction, opening new directions for designing and evaluating LLM systems beyond scaling alone.

Vanessa Figueiredo, Wilter Franceschi

Orbital Sea Regina, SK, Canada

AI Agent ReliabilityLanguage Model-Based ControlLLM ReliabilityLLM Context ManagementLLM Agent Reliability
AskPDF
Feb 21, 2026

Think2^{2}: Grounded Metacognitive Reasoning in Large Language Models

Large Language Models (LLMs) demonstrate strong reasoning performance, yet their ability to reliably monitor, diagnose, and correct their own errors remains limited. We introduce a psychologically grounded metacognitive framework that operationalizes Ann Brown's regulatory cycle (Planning, Monitoring, and Evaluation) as a structured prompting architecture, and study its integration within a lightweight dual-process MetaController for adaptive effort allocation. Across diverse reasoning and diagnostic benchmarks (GSM8K, CRUXEval, MBPP, AIME, CorrectBench, and TruthfulQA) using Llama-3 and Qwen-3 (8B), explicit regulatory structuring substantially improves error diagnosis and yields a threefold increase in successful self-correction. Blinded human evaluations over 580 query pairs show an 84% aggregate preference for trustworthiness and metacognitive self-awareness over standard and Chain-of-Thought baselines. Grounding LLM reasoning in established cognitive theory offers a principled path toward more transparent and diagnostically robust AI systems.

Abraham Paul Elenjical, Vivek Hruday Kavuri, Vasudeva Varma

IIIT Hyderabad

Language Model Self-AssessmentMetacognition in Language ModelsLLM Self-CorrectionLLM ReliabilityAdaptive Computation
AskPDF
Apr 13, 2026

CausalGaze: Unveiling Hallucinations via Counterfactual Graph Intervention in Large Language Models

Despite the groundbreaking advancements made by large language models (LLMs), hallucination remains a critical bottleneck for their deployment in high-stakes domains. Existing classification-based methods mainly rely on static and passive signals from internal states, which often captures the noise and spurious correlations, while overlooking the underlying causal mechanisms. To address this limitation, we shift the paradigm from passive observation to active intervention by introducing CausalGaze, a novel hallucination detection framework based on structural causal models (SCMs). CausalGaze models LLMs' internal states as dynamic causal graphs and employs counterfactual interventions to disentangle causal reasoning paths from incidental noise, thereby enhancing model interpretability. Extensive experiments across four datasets and three widely used LLMs demonstrate the effectiveness of CausalGaze, especially achieving 3.3% improvement in AUROC on the TruthfulQA dataset compared to state-of-the-art baselines.

Linggang Kong, Lei Wu, Yunlong Zhang +4

College of Electronic Engineering, National University of Defense Technology · Anhui Province Key Laboratory of Cyberspace Security Situation Awareness and Evaluation · Institute of Computer Application, China Academy of Engineering Physics

LLM Hallucination DetectionHallucination in Language ModelsCausal Interventions in Language ModelsLLM Interpretability
AskPDF