arXiv subjects

Samar Ansari

Publications and source records attributed to Samar Ansari.

7 recordsLinked to original sources

arXiv2026

From Policing Provenance to Designing for Evidence: Testing Four Propositions on Generative AI Against Research and Institutional Policy

Generative AI has made it hard to establish whether a student produced a piece of assessed work. Building on the argument that validity matters more than cheating, this paper tests four propositions drawn from a strongly stated position: that redesign is more sustainable than detecting or banning AI; that asking students whether and how much AI they used yields low-value information; that unaided ability should be tested directly in controlled conditions; and that the same logic applies to peer review. The propositions are tested against two purposively sampled, coded corpora: 41 research and sector sources and 55 university and sector policy documents. Coding was by a single coder and the policy sample is UK-weighted, so counts describe the sample, not the sector. The propositions hold unevenly. Redesign over detection is well supported, but only as redesign plus assurance: 4 of 55 documents prohibit AI by default, though about 10 of the 55 are publisher or editorial policies that govern authors rather than student assessment; of the 23 that take a position on detection, 16 reject detectors and one requires them. Testing unaided ability is supported as one lane within a programme, not as a return to invigilation. The disclosure proposition is complicated: institutions treat disclosure as valuable, yet the limited evidence suggests many students do not complete required declarations. The peer-review extension is plausible but under-examined. The field is converging on graded permission scales and two-lane designs, and 31 of 55 documents define named categories of use. From these findings the paper builds a nine-step decision chain that extends outcome-first design logic: the role of AI is decided after the evidence an outcome requires. It also separates design moves that add temporary friction from those that provide assurance. A worked example and the full policy matrix are included.

cs.CY↗

arXiv2026

3, 7, 14, 17, 42: Instruction Tuning Installs and Sharpens Number-Choice Heuristics in Language Models

We measure, without sampling error, the probability a language model assigns to each integer from 1 to N when asked for a random number. For a fixed prompt the model defines one probability vector, which we compute by summing over each integer's scored spellings rather than estimate by sampling, removing the noise that limits earlier studies [1, 4]. We evaluate 18 open-weight model lines at five ranges up to N = 100, under 29 decoding configurations, on two inference stacks; 15 of them have matched base and instruction-tuned checkpoints that both give usable answers. Instruction tuning makes the distribution less uniform in all 15 pairs. It installs one heuristic, round-number avoidance, and sharpens another, endpoint avoidance. Base checkpoints do not avoid round numbers once the endpoints are removed from those sets; tuned checkpoints avoid them, multiples of 10 almost completely. Base checkpoints return the endpoints 1 and N at 0.30 to 0.60 of the interior rate; tuned checkpoints almost never return them. A preference for odd numbers is consistent across tuned checkpoints only at N = 100, where base checkpoints also show it weakly. Endpoint avoidance and the odd preference fit the account Kubovy and Psotka [6] proposed for human number choice; round-number avoidance is not part of that account, and the midpoint avoidance it predicts is essentially absent, with one slight exception at N = 100. Tuning does not install a new favourite integer shared across models: above N = 10 no modal integer is shared by more than 42 percent of tuned conditions, and P (17) rises in 10 of 15 pairs and falls in 5. The concentration is present in the model's own distribution: at temperature 1.0, nucleus truncation at top-p 0.95 adds a median of 0.035 to the total variation distance.

cs.CL↗

arXiv2026

Beyond Training: A Feasibility Taxonomy for Inference-Time AI Governance

Compute governance today is a governance of training: the thresholds, reporting requirements, and frontier-AI regimes now in force attach to training compute and treat the trained model as the regulatory unit. That picture is incomplete: capability increasingly migrates to the deployment stage through inference-time scaling, agentic scaffolding, and compression onto consumer hardware. This paper asks which mechanisms are available once the regulatory object shifts from the training run to the inference call. We develop a feasibility taxonomy of twenty inference-time mechanisms across monitoring, verification, and enforcement, each rated on a four-point readiness scale against a documented four-vendor evidence base. We then stress the taxonomy against a two-dimensional adversary model (three capability tiers crossed with four adversary roles) and map each mechanism to four governance scenarios (domestic regulation, bilateral or multilateral coordination, industry self-regulation, and compute-marketplace governance). Fifteen of the twenty mechanisms have commercial technical substrates in production today, although governance-grade assurance and adversarial robustness vary substantially. The adversary analysis shows that this readiness holds only against a cooperative deployer and a low-to-medium-capability user: no mechanism rates adequate against a high-capability state-level deployer, and fine-tuning removes the model-internal components of the enforcement cluster, although platform-external controls can persist. A substitution analysis connects the taxonomy to a companion hardware paper as a conditional substitution principle describing when inference-stage and hardware-stage mechanisms provide comparable regulatory coverage under stated conditions. A second-rater reliability check on a random subset of the readiness ratings returned a quadratic-weighted Cohen's kappa of 0.74.

cs.CY↗

arXiv2026

The Slop Paradox: How Synthetic Standardization Erodes Clinical Uncertainty and Cross-Modal Alignment in AI-Rewritten Radiology Reports

AI-assisted clinical documentation tools increasingly summarize, standardize, and reformat radiology reports using large language models (LLMs). We present a controlled measurement of the resulting information degradation. Using 450 chest X-ray reports from the Indiana University dataset, we generate synthetic versions via three realistic LLM rewriting tasks: EHR summarization, standardized rewriting, and teaching case preparation. We measure entity erosion (via medical NER), hedging collapse (loss of clinical uncertainty language), and cross-modal alignment degradation (via BiomedCLIP image-text similarity). Our central finding is a dissociation between information loss and cross-modal fidelity. EHR summarization is the most destructive at the content level, eroding 51.4% of clinical entities and 43.7% of hedging language, yet it preserves image-text alignment almost entirely (a 2.5% drop). The two tasks meant to produce cleaner training data, standardized rewriting and teaching case preparation, do the reverse: they preserve more entities (26.8% and 29.3% eroded) but cause 14.9-16.5% alignment drops, six to seven times those of EHR summarization. We term this the slop paradox: rewriting that makes clinical text look cleaner for multimodal training is precisely what pulls it away from the image. Contrary to our pre-specified hypothesis, rare pathologies were not preferentially degraded: across nine rare-versus-common comparisons, no difference survived multiple-comparison correction, and nominal differences ran in the opposite direction (common > rare), so contamination is invisible to condition-specific monitoring. The dominant determinant of degradation is the type of AI rewriting task, not the clinical content. These findings bear on multimodal medical AI dataset construction and the governance of AI-assisted clinical documentation.

cs.CL↗

arXiv2026

Hardware-Level Governance of AI Compute: A Feasibility Taxonomy for Regulatory Compliance and Treaty Verification

The governance of frontier AI increasingly relies on controlling access to computational resources, yet the hardware-level mechanisms invoked by policy proposals remain largely unexamined from an engineering perspective. This paper bridges the gap between AI governance and computer engineering by proposing a taxonomy of 20 hardware-level governance mechanisms, organised by function (monitoring, verification, enforcement) and assessed for technical feasibility on a four-point scale from currently deployable to speculative. For each mechanism, we provide a technical description, a feasibility rating, and an identification of adversarial vulnerabilities. We map the taxonomy onto four governance scenarios: domestic regulation, bilateral agreements, multilateral treaty verification, and industry self-regulation. Our analysis reveals a structural mismatch: the mechanisms most needed for treaty verification, including on-chip compute metering, cryptographic proof-of-training, and hardware-embedded enforcement, are also the least mature. We assess principal threats to compute-based governance, including algorithmic efficiency gains, distributed training methods, and sovereignty concerns. We identify a temporal constraint: the window during which semiconductor manufacturing concentration makes hardware-level governance implementable is narrowing, while R&D timelines for critical mechanisms span years. We present an adversary-tiered threat analysis distinguishing commercial, non-state, and nation-state actors, arguing the appropriate security standard is tamper-evident assurance analogous to IAEA verification rather than absolute tamper-proofing. The taxonomy, feasibility classification, and mechanism-to-scenario mapping provide a technical foundation for policymakers and identify the R&D investments required before hardware-level governance can support verifiable international agreements.

cs.CR↗

arXiv2026

PrivHAR-Bench: A Graduated Privacy Benchmark Dataset for Video-Based Action Recognition

Existing research on privacy-preserving Human Activity Recognition (HAR) typically evaluates methods against a binary paradigm: clear video versus a single privacy transformation. This limits cross-method comparability and obscures the nuanced relationship between privacy strength and recognition utility. We introduce \textit{PrivHAR-Bench}, a multi-tier benchmark dataset designed to standardize the evaluation of the \textit{Privacy-Utility Trade-off} in video-based action recognition. PrivHAR-Bench applies a graduated spectrum of visual privacy transformations: from lightweight spatial obfuscation to cryptographic block permutation, to a curated subset of 15 activity classes selected for human articulation diversity. Each of the 1,932 source videos is distributed across 9 parallel tiers of increasing privacy strength, with additional background-removed variants to isolate the contribution of human motion features from contextual scene bias. We provide lossless frame sequences, per-frame bounding boxes, estimated pose keypoints with joint-level confidence scores, standardized group-based train/test splits, and an evaluation toolkit computing recognition accuracy and privacy metrics. Empirical validation using R3D-18 demonstrates a measurable and interpretable degradation curve across tiers, with within-tier accuracy declining from 88.8\% (clear) to 53.5\% (encrypted, background-removed) and cross-domain accuracy collapsing to 4.8\%, establishing PrivHAR-Bench as a controlled benchmark for comparing privacy-preserving HAR methods under standardized conditions. The dataset, generation pipeline, and evaluation code are publicly available.

cs.CV↗

arXiv2026

Compound Deception in Elite Peer Review: A Failure Mode Taxonomy of 100 Fabricated Citations at NeurIPS 2025

Large language models (LLMs) are increasingly used in academic writing workflows, yet they frequently hallucinate by generating citations to sources that do not exist. This study analyzes 100 AI-generated hallucinated citations that appeared in papers accepted by the 2025 Conference on Neural Information Processing Systems (NeurIPS), one of the world's most prestigious AI conferences. Despite review by 3-5 expert researchers per paper, these fabricated citations evaded detection, appearing in 53 published papers (approx. 1% of all accepted papers). We develop a five-category taxonomy that classifies hallucinations by their failure mode: Total Fabrication (66%), Partial Attribute Corruption (27%), Identifier Hijacking (4%), Placeholder Hallucination (2%), and Semantic Hallucination (1%). Our analysis reveals a critical finding: every hallucination (100%) exhibited compound failure modes. The distribution of secondary characteristics was dominated by Semantic Hallucination (63%) and Identifier Hijacking (29%), which often appeared alongside Total Fabrication to create a veneer of plausibility and false verifiability. These compound structures exploit multiple verification heuristics simultaneously, explaining why peer review fails to detect them. The distribution exhibits a bimodal pattern: 92% of contaminated papers contain 1-2 hallucinations (minimal AI use) while 8% contain 4-13 hallucinations (heavy reliance). These findings demonstrate that current peer review processes do not include effective citation verification and that the problem extends beyond NeurIPS to other major conferences, government reports, and professional consulting. We propose mandatory automated citation verification at submission as an implementable solution to prevent fabricated citations from becoming normalized in scientific literature.

cs.DL↗