Research Papers

In recent years, several advances have been observed in Deep Learning with surprising results. Models in this area have been increasingly used in numerous applications, including those sensitive to human life, which require clear explanations and justifications. Various explainability methods have been proposed, but not many metrics to evaluate these methods. The most commonly used metric is the Intersection over Union (IoU). However, due to the characteristics of the results of the explainability methods, called saliency maps, which do not have a known shape, we hypothesise that there must be a better metric that allows one to find an explainability method that produces results that best resemble the human perception. We propose using different metrics to assess the similarity between human perception and the explanation saliency maps to find a better metric. An investigation was conducted employing a subset of the Chihuahuas images from ImageNet dataset. Several CAM-based explainability methods were used to generate saliency maps for each chihuahua image. Alignment was measured by applying distance metrics between the bounding box of human annotations and the saliency maps produced by each explainability method. Rankings of the best saliency maps were created using the results of the distance metrics and compared to the ranking obtained using people's choice, collected through crowdsourcing, of the best explanation saliency maps for each selected image. Comparison between rankings was performed using the Rank-Biased Overlap (RBO) metric. The results indicate the feasibility of our method to find the explainability method that best resembles human perception. In our experiments, the two metrics that best resemble human perception corresponded to Manhattan and Correlation. Besides, the best explainability methods regarding human perception were LayerCAM, Score-CAM, and IS-CAM.

This work establishes a framework for standardizing AI evaluation RCTs (sometimes called human uplift studies). Drawing on established practices from disciplines with established RCT traditions, including software engineering, economics, clinical and health sciences, and psychology, we synthesize five principles drawn from established validity frameworks and open-science standards on transparency, repeatability, and verification, which together serve as the conceptual foundation for 33 actionable guidelines adapted for AI evaluation RCT contexts, expressed as requirements with rationales, implementation instructions, and evidence bases. We position the principles and guidelines as serving three key roles for AI evaluation RCTs: a design tool for planning studies, an evaluation rubric for assessing existing work, and a blueprint for standard setting as the field converges on norms. AI evaluation research currently lacks common standards and shared vocabulary for producing cumulative, comparable, policy-ready evidence. This framework is a contribution toward that foundation, providing evaluative criteria and a shared conceptual language alongside actionable guidelines.

Recent AI systems compress the distance between capability growth and capability deployment. Earlier high-risk technologies were slowed by capital intensity, physical bottlenecks, organizational inertia, and specialized supply chains. By contrast, AI capabilities can be copied, invoked, embedded in workflows, and scaled across institutions at low marginal cost. This paper argues that declining deployment friction changes the safety problem at its root. Safety is not only local output correctness or preference alignment, but the control of irreversibility under rising decision density. The paper formalizes this claim through decision-energy density: the rate-weighted capacity of a node to generate, evaluate, select, and execute consequential decisions. It then identifies three sovereignty boundaries that determine whether AI remains an amplifier within a human-governed system or becomes a de facto control center: irreversible decision authority, physical resource mobilization authority, and self-expansion authority. The model shows how efficiency pressure, path dependence, scale feedback, and weak boundary constraints concentrate decision-energy in the most efficient node. This concentration can diffuse responsibility and raise the probability of irreversible system-level loss even when local per-action error rates remain low. The main result is a boundary stabilization theorem. It shows that safety need not require proving that advanced systems are always correct. Instead, it requires institutional and technical designs that prevent irreversible power from being released by a single high-efficiency node. The paper reframes AI safety as layered control, authorization, and externally reviewable limits, linking alignment, security engineering, organizational economics, and institutional design.

There has been a growing interest in explaining entailments over description logic (DL) knowledge bases. The existing explanation formalisms focus on justifications to explain true axioms, and abductive reasoning to explain missing axioms in a knowledge base. However, these formalisms only point out the reasoning steps behind a (missing) entailment and lack a user-centered approach as they do not consider an inquirer's needs, level of understanding, or prior knowledge. We propose contrastive explanations, aiming at answering "why an axiom P (fact) is true instead of another axiom Q (foil)" over description logic knowledge bases. The motivation arises from the observation that when a user discovers that P has occurred, they are often surprised because they anticipated the occurrence of another similar event Q. Furthermore, individual explanations for "why P" and "why not Q" are unsatisfactory since a user expects to see the difference between P and Q. In this work, we first present formal foundations of contrasting questions and then define contrastive explanations within description logics. To this end, facts include ABox assertions of the form C(x) for a concept C and individual x. Possible foils for such facts are assertions C(y) (contrasting against an individual y), or D(x) (contrasting against a concept D). Additionally, we explore the properties of contrastive explanations in the DL EL and ALC. We also provide an implementation of our definition and an experimental evaluation on KBs of varying sizes.

The Case for ESM3 as a General-Purpose AI Model with Systemic Risk Under the EU AI Act

Taro Qureshi, Jacob Griffith, Koen Holtman +3

Due to ambiguity in the wording of the EU AI Act, we examine the question of to what extent frontier biological foundation models such as ESM3 are subject to obligations for general-purpose AI models with systemic risk under the EU AI Act. In this paper, we map ESM3 to the biorisk chain, and conclude that it would be desirable if the providers of ESM3 and similar biological models were subject to these obligations, which would require them to assess and mitigate dual-use risks from their models. We then perform an analysis, comparing the attributes of ESM3 to the classification criteria in the AI Act and the supporting material. We conclude that at this time, ESM3 does not appear to be meaningfully regulated by the Act. We then propose remedies to correct the situation.

Large language models process the world's writing systems with radical inequality. We constructed the Digital Script Representation Index (DSRI), a seven-axis measure of digital support, and applied it to the 300 writing systems of the Global Script Database (Fukui, 2026). Only 29 scripts (9.7%) are fully supported by contemporary digital infrastructure; among 158 living scripts, 60 (38.0%) lack complete support. Tokenizer efficiency varies by a factor of 31.7 across 45 scripts measured with parallel text. A serial mediation model -- imperial intervention to speaker population to web corpus to tokenizer efficiency -- is consistent with full mediation, with the direct effect of empire indistinguishable from zero (beta = -0.22, p = 0.39) and structural equation model fit indices indistinguishable from saturation at n = 45; the bias-corrected bootstrap CI grazes zero, and we treat the mediation as suggestive rather than confirmatory. Across four independent LLM families (Claude, GPT-4o, Grok, DeepSeek; 12,000 API calls), base-rate-deviation error patterns converge at Spearman rho = 0.85-0.98 (all p < 0.002). 172 script-feature items are answered identically wrong by all four models; over-attribution outnumbers under-recognition 3.9:1, and "used for religion" alone concentrates 43.6% of convergent errors (enrichment 4.1x). With religion excluded as a sensitivity check, the cross-architecture convergence is preserved (mean rho = 0.87 on nine features) and the over-attribution asymmetry persists at 1.77:1 (n = 97, binomial p = 0.008), indicating multi-channeled rather than single-channeled bias. The findings are consistent with an interpretation in which the structural inequalities historical empires inflicted on script communities persist in contemporary language models through the shared training corpus rather than through any individual model's design choices.

The growing reliance of machine learning models in high-stakes, highly regulated domains such as finance and insurance has created a growing tension between predictive performance, interpretability, and regulatory fairness requirements. In these settings, models are expected not only to deliver reliable predictions but also to provide transparent decision rationales and comply with strict fairness requirements. Attention-based transformers offer powerful mechanisms for modeling complex data relationships as demonstrated in various language tasks, yet their attention mechanisms alone do not ensure counterfactually fair predictions, even when combined with fairness-aware techniques. To address these limitations, we propose the Feature Correlation Transformer (FCorrTransformer), an attention-light architecture tailored for tabular data. In this design, the attention matrix admits a direct statistical interpretation as pairwise feature dependencies, enhancing both interpretability and efficiency. Leveraging this structure, we introduce Counterfactual Attention Regularization (CAR), a framework that enforces group-invariant fair representations of sensitive features at the attention level, promoting counterfactually fair predictions without relying on explicit causal assumptions. Empirical evaluations on imbalanced classification and regression benchmarks demonstrate that FCorrTransformer combined with CAR achieves strong counterfactual fairness while maintaining competitive predictive performance and substantially reducing model complexity compared with standard transformer-based baselines. Overall, this work bridges a critical gap between fairness theory and machine learning models, offering a practical framework for responsible AI in regulatory-sensitive domains.

TRUST: A Framework for Decentralized AI Service v.0.1

Yu-Chao Huang, Zhen Tan, Mohan Zhang +3

Large Reasoning Models (LRMs) and Multi-Agent Systems (MAS) in high-stakes domains demand reliable verification, yet centralized approaches suffer four limitations: (1) Robustness, with single points of failure vulnerable to attacks and bias; (2) Scalability, as reasoning complexity creates bottlenecks; (3) Opacity, as hidden auditing erodes trust; and (4) Privacy, as exposed reasoning traces risk model theft. We introduce TRUST (Transparent, Robust, and Unified Services for Trustworthy AI), a decentralized framework with three innovations: (i) Hierarchical Directed Acyclic Graphs (HDAGs) that decompose Chain-of-Thought reasoning into five abstraction levels for parallel distributed auditing; (ii) the DAAN protocol, which projects multi-agent interactions into Causal Interaction Graphs (CIGs) for deterministic root-cause attribution; and (iii) a multi-tier consensus mechanism among computational checkers, LLM evaluators, and human experts with stake-weighted voting that guarantees correctness under 30% adversarial participation. We prove a Safety-Profitability Theorem ensuring honest auditors profit while malicious actors incur losses. All decisions are recorded on-chain, while privacy-by-design segmentation prevents reconstruction of proprietary logic. Across multiple LLMs and benchmarks, TRUST attains 72.4% accuracy (4-18% above baselines) and remains resilient against 20% corruption. DAAN reaches 70% root-cause attribution (vs. 54-63% for standard methods) with 60% token savings. Human studies validate the design (F1 = 0.89, Brier = 0.074). The framework supports (A1) decentralized auditing, (A2) tamper-proof leaderboards, (A3) trustless data annotation, and (A4) governed autonomous agents, pioneering decentralized AI auditing for safe, accountable deployment of reasoning-capable systems.

This article reviews and updates the framework for accountability in AI based on account- ability ecosystems. We update the framework in light of the latest developments since the release of Large Language Models for general public use. We propose three interlinked updates to the original AI accountability ecosystem: (i) reorienting the accountability ecosystem to AI infrastructure and supply chains, (ii) providing greater emphasis on outcomes monitoring and identification of issues that support decentralized system improvement, and (iii) incorporating end-user accountability given the new risks of unpredictability of language models in-the-wild. Collectively, these updates mark a shift towards accountability as distributed, continuous, and institutionalized, away from a system in which frontier AI applications can be modeled as discrete products controlled by single identifiable actors with industry-specific oversight.

People-Centred Medical Image Analysis via Fairness-Aware Human-AI Cooperation

Zheng Zhang, Milad Masroor, Cuong Nguyen +6

Machine learning models for medical image analysis often exhibit subgroup-dependent performance, which impacts how decisions should be allocated between automated systems and human experts under limited resources. Prior work on AI fairness and human-AI cooperation, including learning to defer (L2D) and learning to complement (L2C), typically addresses these problems in isolation. We propose People-Centred Medical Image Analysis (PecMan), a framework for fairness-aware human-AI co-operative classification that jointly models subgroup-dependent reliability, decision allocation, and collaborative prediction. PecMan combines subgroup-specialised predictors with a gating and consolidation mechanism that dynamically assigns cases to automated models, human experts, or their combination, without requiring sensitive attributes at test time. We also introduce the FairHAI benchmark for evaluating trade-offs between predictive accuracy, subgroup equity, and human involvement. In addition, we provide a theoretical analysis of multi-agent gating via selection regret and characterise fairness-coverage trade-offs under input-dependent allocation. Experiments across multiple medical imaging datasets demonstrate that PecMan achieves consistently improved trade-offs compared to methods that address fairness or human-AI cooperation separately.

Green Shielding: A User-Centric Approach Towards Trustworthy AI

Aaron J. Li, Nicolas Sanchez, Hao Huang +8

Large language models (LLMs) are increasingly deployed, yet their outputs can be highly sensitive to routine, non-adversarial variation in how users phrase queries, a gap not well addressed by existing red-teaming efforts. We propose Green Shielding, a user-centric agenda for building evidence-backed deployment guidance by characterizing how benign input variation shifts model behavior. We operationalize this agenda through the CUE criteria: benchmarks with authentic Context, reference standards and metrics that capture true Utility, and perturbations that reflect realistic variations in the Elicitation of model behavior. Guided by the PCS framework and developed with practicing physicians, we instantiate Green Shielding in medical diagnosis through HealthCareMagic-Diagnosis (HCM-Dx), a benchmark of patient-authored queries, together with structured reference diagnosis sets and clinically grounded metrics for evaluating differential diagnosis lists. We also study perturbation regimes that capture routine input variation and show that prompt-level factors shift model behavior along clinically meaningful dimensions. Across multiple frontier LLMs, these shifts trace out Pareto-like tradeoffs. In particular, neutralization, which removes common user-level factors while preserving clinical content, increases plausibility and yields more concise, clinician-like differentials, but reduces coverage of highly likely and safety-critical conditions. Together, these results show that interaction choices can systematically shift task-relevant properties of model outputs and support user-facing guidance for safer deployment in high-stakes domains. Although instantiated here in medical diagnosis, the agenda extends naturally to other decision-support settings and agentic AI systems.

Forecasting when AI systems will become capable of meaningfully accelerating AI research is a central challenge for AI safety. Existing benchmarks measure broad capability growth, but may not provide ample early warning signals for recursive self-improvement. We propose measuring AI's capability to autonomously implement end-to-end machine learning pipelines from past AI research breakthroughs, given a minimal task description. By providing a concise task description instead of the full prior work as reference, we hope to better elicit emerging AI research taste. We introduce a proof-of-concept benchmark in which frontier coding agents autonomously implement an AlphaZero-style machine learning pipeline for Connect Four on consumer hardware within a three-hour budget, and we evaluate the resulting game AIs in a round-robin tournament anchored to the Pascal Pons Connect Four solver. Across four agents with eight trials each, we find substantial differentiation: Claude Opus 4.7 won as first-mover against Pons in seven of eight trials, statistically significantly better than the other agents tested, none of which exceeded two of eight. The task, which no frontier agent could reliably complete when we began development in January of 2026, is now near-saturation. Our evaluation also surfaced anomalous behavior in GPT-5.4, which consistently used far less of its allocated time budget than other agents. A follow-up 16-trial probe using shorter, less evaluation-coded prompts substantially increased GPT-5.4's time-budget usage, consistent with but not diagnostic of sandbagging; Bradley-Terry ratings across probe conditions showed only directional differences, despite significant differences in time-budget usage. We release our data, code, and prompts to support reproduction and extension.