Weekly research digest

AI Safety Digest

Papers, technical posts, and reports on alignment, interpretability, evaluations, and AI governance — collected from 10 organizations every Monday.

August 17, 2026 — August 24, 2026 38 papers this week

Featured this week

Anthropic Aug 24, 2026

Would This Change Your Answer? Evaluating Explanations of LLM Behavior in the Wild with Counterfactual Experiments

We introduce CHIVE, an agentic pipeline that discovers unexpected LLM behaviors in the wild and explains them with counterfactual prompt edits. We use the resulting data in two ways. Using it as an evaluation, we find that activation-reading interpretability tools provide no uplift: agents given the tools predict the outcomes of these experiments no better than agents that just read the transcript. Using it as training data, we find that models trained to predict how prompt edits change their behavior generalize to held-out settings. More
We introduce CHIVE, an agentic pipeline that discovers unexpected LLM behaviors in the wild and explains them with counterfactual prompt edits. We use the resulting data in two ways. Using it as an evaluation, we find that activation-reading interpretability tools provide no uplift: agents given the tools predict the outcomes of these experiments no better than agents that just read the transcript. Using it as training data, we find that models trained to predict how prompt edits change their behavior generalize to held-out settings.
Read paper →
Microsoft Research Aug 20, 2026

Broadening access to Skala creates a faster path to predictive DFT

Sebastian Ehlert, Stefano Battaglia, Thijs Vogels, Jan Hermann, Jens Wehner, Giulia Luise, Klaas Giesbertz, Chin-Wei Huang, Aaron Kaplan, Kate Milton, Stephanie Marisa Lanius, Derk Kooi, P. Bernát Szabó, Gregor Simm, Rianne van den Berg, Paola Gori Giorgi

Skala 1.1, the updated deep-learning exchange-correlation functional from Microsoft Research, provides greater accuracy, expanded accessibility across the computational chemistry ecosystem, and a living benchmark to track computational performance. The post Broadening access to Skala creates a faster path to predictive DFT appeared first on Microsoft Research . More
Skala 1.1, the updated deep-learning exchange-correlation functional from Microsoft Research, provides greater accuracy, expanded accessibility across the computational chemistry ecosystem, and a living benchmark to track computational performance. The post Broadening access to Skala creates a faster path to predictive DFT appeared first on Microsoft Research .
Read paper →

All papers

GovAI Aug 24, 2026

Confronting the Barriers to AI Diffusion in the US Military

AI-enabled command-and-control systems and low-cost, semi-autonomous drones are reshaping battlefields in Ukraine and the Middle East, while China is marshalling its techno-industrial base in an attempt to establish itself as a world-class military power. In the United States, the White House wants to “aggressively adopt” AI throughout the US military, and the Pentagon has directed the Department of Defense (DoD) to become an “AI-first” warfighting force. Whether the United States can maintain its military edge may depend on how fast it builds, adopts, and fields emerging technologies. But while AI is playing a growing role in decision support and intelligence analysis, wider and deeper diffusion could prove more challenging, especially in physical autonomous systems. This paper categorizes and assesses the range of barriers that will slow the US military’s AI transformation, using autonomous drones as a case study. More
AI-enabled command-and-control systems and low-cost, semi-autonomous drones are reshaping battlefields in Ukraine and the Middle East, while China is marshalling its techno-industrial base in an attempt to establish itself as a world-class military power. In the United States, the White House wants to “aggressively adopt” AI throughout the US military, and the Pentagon has directed the Department of Defense (DoD) to become an “AI-first” warfighting force. Whether the United States can maintain its military edge may depend on how fast it builds, adopts, and fields emerging technologies. But while AI is playing a growing role in decision support and intelligence analysis, wider and deeper diffusion could prove more challenging, especially in physical autonomous systems. This paper categorizes and assesses the range of barriers that will slow the US military’s AI transformation, using autonomous drones as a case study.
Read paper →
CSET Aug 24, 2026

Outpaced: AI and Policy’s Role in Transforming Cybersecurity Compliance

Despite decades of reform efforts, the federal government’s Authorization to Operate (ATO) process is a major barrier to delivering secure systems to warfighters at speed. This report examines why past ATO reforms have fallen short of their intent, and offers recommendations to getting leading American technology into the hands of warfighters faster and securely. More
Despite decades of reform efforts, the federal government’s Authorization to Operate (ATO) process is a major barrier to delivering secure systems to warfighters at speed. This report examines why past ATO reforms have fallen short of their intent, and offers recommendations to getting leading American technology into the hands of warfighters faster and securely.
Read paper →
SemiAnalysis Aug 21, 2026

Are Open Models Catching Up?

Evan Cloutier

The past two months have been a breakout period for open source AI. Yes, there was the “DeepSeek moment” back in January 2025, but no one actually used R1 to do any economically valuable work. In contrast, models like GLM 5.3 and Kimi K3 are genuinely capable of many of the same coding and agentic tasks that rocketed Anthropic to $65B+ ARR. Unlike others who inflated ARR, our figures were much closer to reality. More
The past two months have been a breakout period for open source AI. Yes, there was the “DeepSeek moment” back in January 2025, but no one actually used R1 to do any economically valuable work. In contrast, models like GLM 5.3 and Kimi K3 are genuinely capable of many of the same coding and agentic tasks that rocketed Anthropic to $65B+ ARR. Unlike others who inflated ARR, our figures were much closer to reality.
Read paper →
Dean Ball Aug 20, 2026

“These Parchment Barriers”

Dean W. Ball

Dear readers,It’s a pleasure to be back in your inbox. I apologize for the long hiatus in my writing. My first six weeks at OpenAI have been, for many reasons, eventful. I’m pleased to share the launch of the AI Futures blog at OpenAI, which will be the home for the work done by my Strategic Futures team at the company. Despite being on OpenAI’s website, the words below only reflect my views and not the views of OpenAI or any other of its employees; the same will be true for all future posts to this blog.I won’t be the only author of AI Futures, but today’s launch post, on the concentration of power, is from me. Its text is reproduced below for your convenience. This new blog will not replace Hyperdimensional, where I expect to have new essays to share with you soon. More
Dear readers,It’s a pleasure to be back in your inbox. I apologize for the long hiatus in my writing. My first six weeks at OpenAI have been, for many reasons, eventful. I’m pleased to share the launch of the AI Futures blog at OpenAI, which will be the home for the work done by my Strategic Futures team at the company. Despite being on OpenAI’s website, the words below only reflect my views and not the views of OpenAI or any other of its employees; the same will be true for all future posts to this blog.I won’t be the only author of AI Futures, but today’s launch post, on the concentration of power, is from me. Its text is reproduced below for your convenience. This new blog will not replace Hyperdimensional, where I expect to have new essays to share with you soon.
Read paper →
arXiv Aug 21, 2026

VIALS: A Benchmark for Visual Interpretation of Artifacts in the Life Sciences

Elaine Lau, Thanuka Udumulla, Lee Izhaki-Tavor, Francisco Guzmán, Nicholas Magazine, Jonas Mueller

In professional life sciences workflows, scientists routinely interpret visual artifacts (gel blots, microscopy images, plasmid maps, flow cytometry plots, molecular structures, ...) to inform research decisions. We introduce VIALS, a visual question-answering benchmark with 161 such interpretation tasks, spanning the types of artifacts examined throughout experimental workflows in the biotech industry (rather than polished figures from publications and textbooks). While frontier vision-language models can now fluently describe natural images, we find that they are unable to accurately interpret these scientific images, reflecting limitations in domain knowledge and domain-specific visual reasoning capabilities. In contrast, scientists with relevant domain expertise find these visual interpretation tasks straightforward. AI that cannot similarly interpret such images will have limited utility in professional life sciences workflows, where such artifacts are central to how scientists reason, communicate, and make decisions. More
In professional life sciences workflows, scientists routinely interpret visual artifacts (gel blots, microscopy images, plasmid maps, flow cytometry plots, molecular structures, ...) to inform research decisions. We introduce VIALS, a visual question-answering benchmark with 161 such interpretation tasks, spanning the types of artifacts examined throughout experimental workflows in the biotech industry (rather than polished figures from publications and textbooks). While frontier vision-language models can now fluently describe natural images, we find that they are unable to accurately interpret these scientific images, reflecting limitations in domain knowledge and domain-specific visual reasoning capabilities. In contrast, scientists with relevant domain expertise find these visual interpretation tasks straightforward. AI that cannot similarly interpret such images will have limited utility in professional life sciences workflows, where such artifacts are central to how scientists reason, communicate, and make decisions.
Read paper →
arXiv Aug 21, 2026

CLEAR: Continuous Latent Adapter Routing for Utility-Preserving LLM Safety Alignment

Chengxiao Wang, Enyi Jiang, Xiaojing Liao, Sanmi Koyejo

Improving the safety of large language models (LLMs) often comes at the expense of utility, as globally applied safety tuning may affect model responses to both harmful and benign inputs. We propose \textbf{C}ontinuous \textbf{L}at\textbf{E}nt \textbf{A}dapter \textbf{R}outing (CLEAR), a conditional safety adaptation framework that uses a lightweight hidden-state gate to continuously control the activation strength of a safety low-rank adapter. CLEAR aims to reduce harmful completions while avoiding unnecessary changes to the frozen backbone that could degrade performance on benign prompts. Experiments on widely used safety and utility benchmarks show that CLEAR improves robustness on HarmBench while reducing the utility degradation observed with globally applied safety tuning such as SFT or standard low-rank adaptation (LoRA). On Llama-3-8B-Instruct, CLEAR reduces HarmBench ASR from 32.3\% to 0.5\%, while retaining most of the base model's utility and achieving up to 7.1 percentage points higher GSM8K accuracy than globally applied SFT or LoRA. More
Improving the safety of large language models (LLMs) often comes at the expense of utility, as globally applied safety tuning may affect model responses to both harmful and benign inputs. We propose \textbf{C}ontinuous \textbf{L}at\textbf{E}nt \textbf{A}dapter \textbf{R}outing (CLEAR), a conditional safety adaptation framework that uses a lightweight hidden-state gate to continuously control the activation strength of a safety low-rank adapter. CLEAR aims to reduce harmful completions while avoiding unnecessary changes to the frozen backbone that could degrade performance on benign prompts. Experiments on widely used safety and utility benchmarks show that CLEAR improves robustness on HarmBench while reducing the utility degradation observed with globally applied safety tuning such as SFT or standard low-rank adaptation (LoRA). On Llama-3-8B-Instruct, CLEAR reduces HarmBench ASR from 32.3\% to 0.5\%, while retaining most of the base model's utility and achieving up to 7.1 percentage points higher GSM8K accuracy than globally applied SFT or LoRA.
Read paper →
arXiv Aug 21, 2026

ConceptTS: LLM-Guided Concept Bottlenecks for Interpretable Multivariate Time-Series Forecasting

Yichen Jiang, Yueqiao Chen, Dongyu Liu

State-of-the-art multivariate time-series forecasters can model complex temporal and cross-variable dependencies, yet their opaque representations provide limited insight into why a particular forecast is produced. This lack of transparency restricts their use in settings where practitioners must understand and assess the factors underlying a prediction. We introduce ConceptTS, an interpretable forecasting framework that organizes its predictions around named, human-readable concepts. ConceptTS uses a large language model to propose task-relevant concepts and generate executable labeling rules, translating the language model's domain knowledge into direct supervision without costly manual concept annotation. The proposed concepts are organized into three complementary bottlenecks that describe the historical context, local forecast intervals, and the full forecast horizon. A shared decoder combines representations derived from their predicted activations to construct the forecast, making the model's decision process explicit and supporting direct concept-level interventions. More
State-of-the-art multivariate time-series forecasters can model complex temporal and cross-variable dependencies, yet their opaque representations provide limited insight into why a particular forecast is produced. This lack of transparency restricts their use in settings where practitioners must understand and assess the factors underlying a prediction. We introduce ConceptTS, an interpretable forecasting framework that organizes its predictions around named, human-readable concepts. ConceptTS uses a large language model to propose task-relevant concepts and generate executable labeling rules, translating the language model's domain knowledge into direct supervision without costly manual concept annotation. The proposed concepts are organized into three complementary bottlenecks that describe the historical context, local forecast intervals, and the full forecast horizon. A shared decoder combines representations derived from their predicted activations to construct the forecast, making the model's decision process explicit and supporting direct concept-level interventions.
Read paper →
arXiv Aug 21, 2026

ReFrame: Evidence-Guided Test-Time Safety Alignment in Multimodal Large Language Models

Wenzheng Jiang, Xuankun Rong, Yuanzhao Zhai, Dawei Feng, Huaimin Wang

While multimodal large language models (MLLMs) extend model capabilities beyond text, they also make safety alignment increasingly challenging. Multimodal safety alignment methods must address cross-modal jailbreaks, safety-awareness failures, and over-sensitive refusals. However, existing methods often rely on retraining or internal-state inspection, limiting their applicability to deployed closed-source MLLMs and motivating test-time safety alignment. We analyze this setting and identify two key obstacles, utility dominance and reasoning inertia, which cause models to overlook latent risks or follow malicious reasoning trajectories. Guided by these insights, we propose ReFrame, a training-free multimodal input reframing framework where two agents share a lightweight locally deployed MLLM: the evidence-generation agent constructs complementary risk and utility evidence, and the rewrite-and-routing agent converts it into a safe proxy prompt and image-routing decision before calling the downstream MLLM, without modifying it or accessing its internal information. More
While multimodal large language models (MLLMs) extend model capabilities beyond text, they also make safety alignment increasingly challenging. Multimodal safety alignment methods must address cross-modal jailbreaks, safety-awareness failures, and over-sensitive refusals. However, existing methods often rely on retraining or internal-state inspection, limiting their applicability to deployed closed-source MLLMs and motivating test-time safety alignment. We analyze this setting and identify two key obstacles, utility dominance and reasoning inertia, which cause models to overlook latent risks or follow malicious reasoning trajectories. Guided by these insights, we propose ReFrame, a training-free multimodal input reframing framework where two agents share a lightweight locally deployed MLLM: the evidence-generation agent constructs complementary risk and utility evidence, and the rewrite-and-routing agent converts it into a safe proxy prompt and image-routing decision before calling the downstream MLLM, without modifying it or accessing its internal information.
Read paper →
arXiv Aug 21, 2026

Designing a Robust LLM-Based Evaluation System for Agentic AI in Drug Discovery Through Human Alignment

Emma Granqvist, Rocío Mercado, Samuel Genheden

Agentic large language model (LLM) systems are reshaping scientific workflows in chemistry and drug discovery, but evaluating their open-ended, tool-augmented outputs remains a fundamental bottleneck. Reference-based metrics such as BLEU and ROUGE fail to capture semantic correctness, while expert human evaluation does not scale to the iteration speed these systems demand. The LLM-as-a-Judge paradigm has emerged as a scalable alternative, but existing drug discovery benchmarks deploy LLM judges without validating their alignment with human experts. In this work, we present an LLM-as-a-Judge evaluation framework for ChatInvent, an agentic drug discovery assistant deployed at AstraZeneca, with four contributions. First, we define four output-quality evaluation dimensions---Completeness, Relevancy, Structural Clarity, and Scope Adherence---alongside deterministic Tool Call Correctness checks. Second, we validate the judge through a human alignment study with five expert annotators, comparing Gemini 3.1 Pro, Claude Opus 4.7, GPT-5, and Llama 3.1 70B as candidate judges. More
Agentic large language model (LLM) systems are reshaping scientific workflows in chemistry and drug discovery, but evaluating their open-ended, tool-augmented outputs remains a fundamental bottleneck. Reference-based metrics such as BLEU and ROUGE fail to capture semantic correctness, while expert human evaluation does not scale to the iteration speed these systems demand. The LLM-as-a-Judge paradigm has emerged as a scalable alternative, but existing drug discovery benchmarks deploy LLM judges without validating their alignment with human experts. In this work, we present an LLM-as-a-Judge evaluation framework for ChatInvent, an agentic drug discovery assistant deployed at AstraZeneca, with four contributions. First, we define four output-quality evaluation dimensions---Completeness, Relevancy, Structural Clarity, and Scope Adherence---alongside deterministic Tool Call Correctness checks. Second, we validate the judge through a human alignment study with five expert annotators, comparing Gemini 3.1 Pro, Claude Opus 4.7, GPT-5, and Llama 3.1 70B as candidate judges.
Read paper →
arXiv Aug 21, 2026

CoST: Semantic-Aware Urban Understanding via Spatial-Temporal Alignment

Yutian Jiang, Jiabo Liu, Xixuan Hao, Yuxuan Liang

Geospatial representation learning from satellite imagery is a fundamental problem for large-scale urban analysis and real-world applications. Despite recent advances, current methods struggle with cross-region generalization and semantic interpretability due to their reliance on region-specific auxiliary data and the neglect of semantic alignment within multi-temporal urban imagery. Therefore, we present CoST, a novel \underline{Co}ntrastive-based \underline{S}patial-\underline{T}emporal framework that aligns spatial context with multi-temporal semantics to extract universal geographic regularities shared across regions. Specifically, CoST explicitly models spatial correlations to capture transferable geographic structures and exploits multi-year urban change semantics to align learned representations with high-level geo-semantics. Extensive experiments demonstrate that CoST consistently achieves superior performance across various downstream tasks and in unseen scenario, yielding an average relative gain of 8.7\% over the strongest competing methods across eight city-indicator settings. The code is available in \href{https://github.com/Arandinglv/CoST}{this repo}. More
Geospatial representation learning from satellite imagery is a fundamental problem for large-scale urban analysis and real-world applications. Despite recent advances, current methods struggle with cross-region generalization and semantic interpretability due to their reliance on region-specific auxiliary data and the neglect of semantic alignment within multi-temporal urban imagery. Therefore, we present CoST, a novel \underline{Co}ntrastive-based \underline{S}patial-\underline{T}emporal framework that aligns spatial context with multi-temporal semantics to extract universal geographic regularities shared across regions. Specifically, CoST explicitly models spatial correlations to capture transferable geographic structures and exploits multi-year urban change semantics to align learned representations with high-level geo-semantics. Extensive experiments demonstrate that CoST consistently achieves superior performance across various downstream tasks and in unseen scenario, yielding an average relative gain of 8.7\% over the strongest competing methods across eight city-indicator settings. The code is available in \href{https://github.com/Arandinglv/CoST}{this repo}.
Read paper →
arXiv Aug 21, 2026

Trojaning the Alignment: Stealthy Backdoor Attacks against Graph Foundation Models

Minhua Lin, Zhicheng Gao, Yilong Wang, Hanqing Lu, Xiang Zhang, Suhang Wang

Graph Foundation Models (GFMs) on text-attributed graphs (TAGs) align graph representations with language semantics to support transferable graph learning. Despite these advantages, the backdoor vulnerability of GFMs on TAGs remains insufficiently understood, especially under graph-language alignment, where graph and text representations are trained to constrain each other in a shared semantic space. Existing backdoor attacks mainly target either the graph side or the text side, treating the two modalities independently. This makes direct adaptation ineffective: graph-only triggers can be constrained by clean text semantics, while text-only triggers alter the language view but do not directly shift the graph representation being aligned and scored. TAGs also impose a stealth challenge because triggers are exposed as both node text and local graph structure, making incoherent trigger attributes or anomalous subgraphs easy to inspect or filter. More
Graph Foundation Models (GFMs) on text-attributed graphs (TAGs) align graph representations with language semantics to support transferable graph learning. Despite these advantages, the backdoor vulnerability of GFMs on TAGs remains insufficiently understood, especially under graph-language alignment, where graph and text representations are trained to constrain each other in a shared semantic space. Existing backdoor attacks mainly target either the graph side or the text side, treating the two modalities independently. This makes direct adaptation ineffective: graph-only triggers can be constrained by clean text semantics, while text-only triggers alter the language view but do not directly shift the graph representation being aligned and scored. TAGs also impose a stealth challenge because triggers are exposed as both node text and local graph structure, making incoherent trigger attributes or anomalous subgraphs easy to inspect or filter.
Read paper →
arXiv Aug 21, 2026

RAG Deserves an Index: Why Ingest-Time Compilation Beats Query-Time Interpretation

Kyle Wild, Yusuke Takahashi, Asako Uraki

Nearly every retrieval-augmented question-answering system in production ships with a hidden interpreter: on each query a language model re-derives the meaning of raw corpus text and then throws that work away. Cheaper models do not close the gap: per-token prices have fallen by orders of magnitude while inference spend has risen, because context volume grows faster than prices fall. This is the modern equivalent of the full-table scan, and the remedy is the one databases found fifty years ago: do the expensive work once, at write time, into a maintained structure that makes reads cheap. A corpus whose read pattern is known before it ever meets a user can and should be indexed too. More
Nearly every retrieval-augmented question-answering system in production ships with a hidden interpreter: on each query a language model re-derives the meaning of raw corpus text and then throws that work away. Cheaper models do not close the gap: per-token prices have fallen by orders of magnitude while inference spend has risen, because context volume grows faster than prices fall. This is the modern equivalent of the full-table scan, and the remedy is the one databases found fifty years ago: do the expensive work once, at write time, into a maintained structure that makes reads cheap. A corpus whose read pattern is known before it ever meets a user can and should be indexed too.
Read paper →
arXiv Aug 21, 2026

Fuzzy-MoE: Interpretable Regime-Conditioned Expert Routing for Non-Stationary Multivariate Time Series Forecasting

Lan Guo, Jie Xiao, Zhao Su, Jun Shen, Haoran Li, Weixia Ma, Qingguo Zhou, Binbin Yong

In non-stationary multivariate time series, different variables and samples often exhibit heterogeneous latent dynamic states, while existing deep forecasting models usually compress them into a unified end-to-end mapping, leading to suboptimal modeling of time-varying dynamics and limited interpretability regarding which forecasting mechanism is activated under different latent states. To overcome these limitations, we reformulate time series forecasting as a unified framework of latent temporal state identification and interpretable expert routing, and propose Fuzzy-MoE, a fuzzy logic-based dynamic Mixture-of-Experts model. Fuzzy-MoE consists of multiple parallel expert mapping networks and a dual-view fuzzy router. By jointly exploiting local convolutional dynamics and global segmented statistics, the router infers latent temporal states and computes expert activation strengths through learnable Gaussian membership functions, enabling explicit IF-THEN rule-based expert selection. This fine-grained routing strategy allows different variables within the same sequence to activate different experts, effectively capturing heterogeneous temporal dynamics while improving model interpretability. More
In non-stationary multivariate time series, different variables and samples often exhibit heterogeneous latent dynamic states, while existing deep forecasting models usually compress them into a unified end-to-end mapping, leading to suboptimal modeling of time-varying dynamics and limited interpretability regarding which forecasting mechanism is activated under different latent states. To overcome these limitations, we reformulate time series forecasting as a unified framework of latent temporal state identification and interpretable expert routing, and propose Fuzzy-MoE, a fuzzy logic-based dynamic Mixture-of-Experts model. Fuzzy-MoE consists of multiple parallel expert mapping networks and a dual-view fuzzy router. By jointly exploiting local convolutional dynamics and global segmented statistics, the router infers latent temporal states and computes expert activation strengths through learnable Gaussian membership functions, enabling explicit IF-THEN rule-based expert selection. This fine-grained routing strategy allows different variables within the same sequence to activate different experts, effectively capturing heterogeneous temporal dynamics while improving model interpretability.
Read paper →
arXiv Aug 20, 2026

An integrated diffusion-weighted imaging processing and interpretation platform for MR-guided radiotherapy

Yunxiang Li, Yan Dai, Yen-Peng Liao, Jie Deng, Jill B De Vis, You Zhang

Background: Magnetic resonance imaging-guided linear accelerators (MR-Linacs) allow diffusion-weighted imaging (DWI) to be acquired at every treatment fraction, but converting these low-signal-to-noise-ratio acquisitions into clinical decisions requires both reliable quantitative processing and an interpretation that reconciles a scattered and often contradictory literature. Purpose: To describe and evaluate an integrated, web-based platform that carries raw MR-Linac DWI to a structured, literature-grounded clinical interpretation, and to assess its retrieval-augmented generation (RAG) interpretation module by independent expert rating. Methods: The platform couples a deep-learning processing pipeline, comprising distortion correction, denoising, and intravoxel incoherent motion (IVIM)/apparent diffusion coefficient (ADC) fitting, with longitudinal region-of-interest analysis and a RAG interpretation agent. The agent reasons over a two-layer knowledge base of curated publications (a structured catalog index plus line-indexed full text), delegates arithmetic to deterministic tools, and is designed to trace each statement to a source document, section, and line range. More
Background: Magnetic resonance imaging-guided linear accelerators (MR-Linacs) allow diffusion-weighted imaging (DWI) to be acquired at every treatment fraction, but converting these low-signal-to-noise-ratio acquisitions into clinical decisions requires both reliable quantitative processing and an interpretation that reconciles a scattered and often contradictory literature. Purpose: To describe and evaluate an integrated, web-based platform that carries raw MR-Linac DWI to a structured, literature-grounded clinical interpretation, and to assess its retrieval-augmented generation (RAG) interpretation module by independent expert rating. Methods: The platform couples a deep-learning processing pipeline, comprising distortion correction, denoising, and intravoxel incoherent motion (IVIM)/apparent diffusion coefficient (ADC) fitting, with longitudinal region-of-interest analysis and a RAG interpretation agent. The agent reasons over a two-layer knowledge base of curated publications (a structured catalog index plus line-indexed full text), delegates arithmetic to deterministic tools, and is designed to trace each statement to a source document, section, and line range.
Read paper →
arXiv Aug 20, 2026

Inject, Align, Recover: Staged Post-Training for Retrieval-Free Document Knowledge Internalization

Qian Kou, Xiaofeng Shi, Xiaosong Qiu, Hua Zhou

Large language models often fail to answer questions about a bounded document collection when the source documents are not retrieved at inference time. We study this setting as document knowledge internalization: converting a fixed corpus into usable parametric knowledge for retrieval-free question answering. We propose IAR (Inject, Align, and Recover), a three-stage post-training framework that separates structured document knowledge injection, QA behavior alignment, and general ability recovery. Unlike conventional continued pretraining, Inject converts source documents into continuation, rewrite, and instruction-conditioned reconstruction objectives. Align then adapts the injected model with answer-only QA supervision, while Recover merges the domain-adapted model with the base instruction model to recover general capabilities. Across Common Corpus (CC) and CCI, and across Llama, Phi, Qwen, and SmolLM model families, IAR improves the domain-primary domain-general frontier for retrieval-free document internalization. More
Large language models often fail to answer questions about a bounded document collection when the source documents are not retrieved at inference time. We study this setting as document knowledge internalization: converting a fixed corpus into usable parametric knowledge for retrieval-free question answering. We propose IAR (Inject, Align, and Recover), a three-stage post-training framework that separates structured document knowledge injection, QA behavior alignment, and general ability recovery. Unlike conventional continued pretraining, Inject converts source documents into continuation, rewrite, and instruction-conditioned reconstruction objectives. Align then adapts the injected model with answer-only QA supervision, while Recover merges the domain-adapted model with the base instruction model to recover general capabilities. Across Common Corpus (CC) and CCI, and across Llama, Phi, Qwen, and SmolLM model families, IAR improves the domain-primary domain-general frontier for retrieval-free document internalization.
Read paper →
arXiv Aug 20, 2026

SAE-Xplainers: Rule-Based Feature Interpretation for Extreme Earth Events

Hugo Porta, Emanuele Dalsasso, Chang Xu, Theo Gnassounou, Devis Tuia

The emergence of large-scale Weather and Climate (W&C) datasets offers new opportunities for modeling extreme Earth events (ExEE) and their impacts using deep learning. However, their adoption in operational settings remains limited by the lack of models' interpretability. While for conventional text and image modalities, tools such as Sparse Autoencoders (SAEs) have proven effective for extracting human-understandable concepts, their use for the analysis of ExEE remains challenging due to the nature of W&C data. To address this, we introduce (i) a geographic location-based modulation of the inputs of SAE to capture the local semantic meaning of environmental patterns, and (ii) an ensemble of rule-based SAE-Xplainers to interpret the resulting high-dimensional features derived from complex, multi-modal environmental predictors. We evaluate our method on three ExEE types: the prediction of fires, and the detection of tropical cyclones and atmospheric rivers. More
The emergence of large-scale Weather and Climate (W&C) datasets offers new opportunities for modeling extreme Earth events (ExEE) and their impacts using deep learning. However, their adoption in operational settings remains limited by the lack of models' interpretability. While for conventional text and image modalities, tools such as Sparse Autoencoders (SAEs) have proven effective for extracting human-understandable concepts, their use for the analysis of ExEE remains challenging due to the nature of W&C data. To address this, we introduce (i) a geographic location-based modulation of the inputs of SAE to capture the local semantic meaning of environmental patterns, and (ii) an ensemble of rule-based SAE-Xplainers to interpret the resulting high-dimensional features derived from complex, multi-modal environmental predictors. We evaluate our method on three ExEE types: the prediction of fires, and the detection of tropical cyclones and atmospheric rivers.
Read paper →
arXiv Aug 20, 2026

DecoVAE: a Lightweight Interpretable Trend-Seasonal VAE Framework for Efficient Probabilistic Time Series Forecasting

Alexander Marusov, Dmitry Anikin, Alexey Zaytsev

Probabilistic time series forecasting remains challenging, largely because modeling distinct trend and seasonal dynamics requires specialized approaches. Existing methods often fail to capture the unique inner properties of these components, lack interpretability, or suffer from heavy memory and runtime overhead. To address these limitations, we propose DecoVAE, a lightweight interpretable trend-seasonal VAE framework that explicitly decomposes time series into trend and seasonal components by applying domain-specific inductive biases. The trend stream enforces structural smoothness using a differential regularizer on the latent trajectory, analogous to the Hodrick-Prescott filter. Concurrently, the seasonal stream operates in the frequency domain via a complex Gaussian VAE, natively capturing the amplitude and phase of periodic patterns. Extensive evaluations across seven real-world benchmarks show that DecoVAE consistently outperforms strong baselines. It achieves reductions of up to 14.96\% in CRPS and 23.30\% in NMAE for short-term forecasting, and up to 52.68\% and 26.51\% for long-term horizons. More
Probabilistic time series forecasting remains challenging, largely because modeling distinct trend and seasonal dynamics requires specialized approaches. Existing methods often fail to capture the unique inner properties of these components, lack interpretability, or suffer from heavy memory and runtime overhead. To address these limitations, we propose DecoVAE, a lightweight interpretable trend-seasonal VAE framework that explicitly decomposes time series into trend and seasonal components by applying domain-specific inductive biases. The trend stream enforces structural smoothness using a differential regularizer on the latent trajectory, analogous to the Hodrick-Prescott filter. Concurrently, the seasonal stream operates in the frequency domain via a complex Gaussian VAE, natively capturing the amplitude and phase of periodic patterns. Extensive evaluations across seven real-world benchmarks show that DecoVAE consistently outperforms strong baselines. It achieves reductions of up to 14.96\% in CRPS and 23.30\% in NMAE for short-term forecasting, and up to 52.68\% and 26.51\% for long-term horizons.
Read paper →
arXiv Aug 20, 2026

Scale-Aware Pretraining of Time Series Foundation Models via Multi-Patch Token Alignment and Hybrid Masking

Taihua Chen, Xiang Ma, Yixin Zhang, Tailin Zhan, Manyu Sun, Lizhen Cui

Pretraining time series foundation models across heterogeneous datasets necessitates effective handling of varying sampling frequencies. Current methods either employ dataset-specific patch sizes and separate FFNs, leading to fragmented representations, or enforce a fixed patch size that neglects inherent temporal variations. To address this, we propose SATS, featuring a scale-aware token alignment mechanism that treats patch size as an explicit notion of scale. By incorporating a contrastive-inspired alignment regularizer, SATS aligns representation spaces across scales while preserving distinct modeling capacities. Furthermore, a hybrid masking strategy combining random and contiguous masking is introduced to capture multi-scale temporal structures. Experimental results on LSTF benchmarks demonstrate that SATS achieves a 9.2% improvement in MSE and an 8.3% gain in GIFT-Eval MASE compared to competitive baselines. Notably, SATS consistently delivers SOTA performance while achieving a 65.6% increase in model efficiency over advanced baselines, highlighting its effectiveness and scalability in time series pretraining. More
Pretraining time series foundation models across heterogeneous datasets necessitates effective handling of varying sampling frequencies. Current methods either employ dataset-specific patch sizes and separate FFNs, leading to fragmented representations, or enforce a fixed patch size that neglects inherent temporal variations. To address this, we propose SATS, featuring a scale-aware token alignment mechanism that treats patch size as an explicit notion of scale. By incorporating a contrastive-inspired alignment regularizer, SATS aligns representation spaces across scales while preserving distinct modeling capacities. Furthermore, a hybrid masking strategy combining random and contiguous masking is introduced to capture multi-scale temporal structures. Experimental results on LSTF benchmarks demonstrate that SATS achieves a 9.2% improvement in MSE and an 8.3% gain in GIFT-Eval MASE compared to competitive baselines. Notably, SATS consistently delivers SOTA performance while achieving a 65.6% increase in model efficiency over advanced baselines, highlighting its effectiveness and scalability in time series pretraining.
Read paper →
arXiv Aug 20, 2026

Towards Clinically Faithful Medical Image Captioning via Enhanced Vision-Language Alignment

Yunseo Lee, Hyun Jun Kim, Heeseung Shin, Changwon Lim

Medical image captioning is a technique that accelerates early-stage diagnostic workflows and enhances the interpretability of medical diagnostic AI systems. However, unlike general image captioning, clinically reliable captioning remains challenging due to grayscale-based modalities, subtle anatomical cues, specialized medical phrasing, and variations in data quality. Despite recent advances in large vision-language models, fluent outputs do not necessarily guarantee sufficient alignment with clinical concept spaces or evaluation criteria. To address this issue, we propose a framework that strengthens clinical alignment by separating and enhancing training-time alignment and inference-time alignment. We build a medical image captioning pipeline that integrates single/dual vision encoders based on BioMedCLIP and SigLIP2, a Q-Former, and a LLaMA-based decoder, and examine the contribution of auxiliary learning for UMLS concept/type prediction. More
Medical image captioning is a technique that accelerates early-stage diagnostic workflows and enhances the interpretability of medical diagnostic AI systems. However, unlike general image captioning, clinically reliable captioning remains challenging due to grayscale-based modalities, subtle anatomical cues, specialized medical phrasing, and variations in data quality. Despite recent advances in large vision-language models, fluent outputs do not necessarily guarantee sufficient alignment with clinical concept spaces or evaluation criteria. To address this issue, we propose a framework that strengthens clinical alignment by separating and enhancing training-time alignment and inference-time alignment. We build a medical image captioning pipeline that integrates single/dual vision encoders based on BioMedCLIP and SigLIP2, a Q-Former, and a LLaMA-based decoder, and examine the contribution of auxiliary learning for UMLS concept/type prediction.
Read paper →
arXiv Aug 20, 2026

LoRA-GA$^2$: Low Rank Adaptation with Multi-step Gradient Adaptive Alignment

Haonan He, Xinyue Fan

Low-Rank Adaptation (LoRA) is a prominent fine-tuning method for large models, achieving competitive performance with reduced memory overhead. However, a persistent performance gap remains between LoRA and full fine-tuning. Recent studies have sought to narrow this gap by employing one-step gradient approximations of pretrained weights to align LoRA updates with the principal directions or intrinsic dimensionalities of full fine-tuning updates. Nevertheless, these approaches fail to capture the full dynamics of the gradients. In this paper, we propose LoRA-GA$^2$, an effective fine-tuning algorithm that fully leverages multi-step gradient information. Specifically, we introduce a lightweight probe for multi-step gradients of pretrained weights that incurs no additional GPU memory cost and only marginal time overhead. We further employ a spectrum-aware, importance-based rank allocation and optimal initialization derived from multi-step gradients. Extensive experimental results demonstrate that LoRA-GA$^2$ consistently outperforms existing LoRA variants while preserving the efficiency advantages of vanilla LoRA. More
Low-Rank Adaptation (LoRA) is a prominent fine-tuning method for large models, achieving competitive performance with reduced memory overhead. However, a persistent performance gap remains between LoRA and full fine-tuning. Recent studies have sought to narrow this gap by employing one-step gradient approximations of pretrained weights to align LoRA updates with the principal directions or intrinsic dimensionalities of full fine-tuning updates. Nevertheless, these approaches fail to capture the full dynamics of the gradients. In this paper, we propose LoRA-GA$^2$, an effective fine-tuning algorithm that fully leverages multi-step gradient information. Specifically, we introduce a lightweight probe for multi-step gradients of pretrained weights that incurs no additional GPU memory cost and only marginal time overhead. We further employ a spectrum-aware, importance-based rank allocation and optimal initialization derived from multi-step gradients. Extensive experimental results demonstrate that LoRA-GA$^2$ consistently outperforms existing LoRA variants while preserving the efficiency advantages of vanilla LoRA.
Read paper →
arXiv Aug 20, 2026

PEA-DPO: Perception-Enhanced Alignment Direct Preference Optimization for MLLMs Alignment

Jiawei Feng, Jiancan Wu, Xingyu Zhu, Junkang Wu, Xiang Wang, Xiangnan He

Direct Preference Optimization (DPO) has emerged as an effective approach for aligning large language models (LLMs) with human preferences. However, its adaptation to multimodal settings remains unexplored. Through representational analysis, we identify a key limitation in multimodal preference optimization, which we term visual insensitivity: models often fail to distinguish between images and those with critical visual context removed. Our theoretical analysis further uncovers two manifestations of this problem, namely Across-Image Insensitivity and Within-Image Insensitivity. To address these challenges, we propose Perception-Enhanced Alignment DPO (PEA-DPO), a framework for multimodal LLMs alignment, which explicitly leverages visual preference signals to overcome visual insensitivity. We further provide a theoretical analysis demonstrating that PEA-DPO provably mitigates both failure modes. Empirical results demonstrate that PEA-DPO enhances sensitivity to visual context while preserving the language modeling capacity of the base model. More
Direct Preference Optimization (DPO) has emerged as an effective approach for aligning large language models (LLMs) with human preferences. However, its adaptation to multimodal settings remains unexplored. Through representational analysis, we identify a key limitation in multimodal preference optimization, which we term visual insensitivity: models often fail to distinguish between images and those with critical visual context removed. Our theoretical analysis further uncovers two manifestations of this problem, namely Across-Image Insensitivity and Within-Image Insensitivity. To address these challenges, we propose Perception-Enhanced Alignment DPO (PEA-DPO), a framework for multimodal LLMs alignment, which explicitly leverages visual preference signals to overcome visual insensitivity. We further provide a theoretical analysis demonstrating that PEA-DPO provably mitigates both failure modes. Empirical results demonstrate that PEA-DPO enhances sensitivity to visual context while preserving the language modeling capacity of the base model.
Read paper →
arXiv Aug 19, 2026

Mechanistic Tomography: Designed Measurement for Control-Oriented Interpretability

Vijay Erramilli

Mechanistic interpretability seeks quantities that models do not expose directly: represented states, component effects, interactions, and responses to interventions. Patching, gradients, Hessian-vector products, and subset interventions provide different measurements under different access assumptions and may target different quantities. We formulate their shared measurement structure as mechanistic tomography: designed measurement for recovering internal mechanisms and intervention effects. For a chosen basis and intervention family, measurements take the form y = Ax + w, where A describes the interventions, x is the target map, and w contains nonlinear response, sampling error, and basis misspecification. This language gives a practical procedure: start with the least costly measurements, test on held-out interventions at the intended scale, calibrate simple mismatch, and expand the measurement family when structured residuals remain. Control provides a demanding validation setting because an estimate that guides an intervention acts as an observer. More
Mechanistic interpretability seeks quantities that models do not expose directly: represented states, component effects, interactions, and responses to interventions. Patching, gradients, Hessian-vector products, and subset interventions provide different measurements under different access assumptions and may target different quantities. We formulate their shared measurement structure as mechanistic tomography: designed measurement for recovering internal mechanisms and intervention effects. For a chosen basis and intervention family, measurements take the form y = Ax + w, where A describes the interventions, x is the target map, and w contains nonlinear response, sampling error, and basis misspecification. This language gives a practical procedure: start with the least costly measurements, test on held-out interventions at the intended scale, calibrate simple mismatch, and expand the measurement family when structured residuals remain. Control provides a demanding validation setting because an estimate that guides an intervention acts as an observer.
Read paper →
arXiv Aug 19, 2026

Graphical Design of Interpretable Architectures

Pietro Barbiero

Designing, implementing, and comparing interpretable architectures requires a formal language to represent them. The most common representations fall short in one of two ways. Symbolic equations give no global view of an architecture at a glance. Probabilistic graphical models and flowcharts do not describe actual tensor manipulations, thus hiding key insights and limiting reproducibility. To close this gap, we introduce a graphical notation for designing interpretable AI architectures, adapted from Penrose tensor notation. This graphical notation gives a global view of an architecture and maps one to one onto PyTorch einsum code. We first use this notation to describe architectures that are interpretable by construction, including concept bottlenecks, sparse probes, prototype networks, neural additive models, and mixtures of linear models. We then diagram the key architectural components of Steerling-8B, a frontier interpretable language model. More
Designing, implementing, and comparing interpretable architectures requires a formal language to represent them. The most common representations fall short in one of two ways. Symbolic equations give no global view of an architecture at a glance. Probabilistic graphical models and flowcharts do not describe actual tensor manipulations, thus hiding key insights and limiting reproducibility. To close this gap, we introduce a graphical notation for designing interpretable AI architectures, adapted from Penrose tensor notation. This graphical notation gives a global view of an architecture and maps one to one onto PyTorch einsum code. We first use this notation to describe architectures that are interpretable by construction, including concept bottlenecks, sparse probes, prototype networks, neural additive models, and mixtures of linear models. We then diagram the key architectural components of Steerling-8B, a frontier interpretable language model.
Read paper →
Alignment Forum Aug 19, 2026

Debate Training Reduces Reward Hacking in RLAIF

zac_kenton

Paper: Debate Training Reduces Reward Hacking in RLAIF Linkpost for GDM Alignment blogpost Work done by the GDM Amplified Oversight team ( we're hiring ). TL;DR : When you RL against an LLM judge, the judge gets hacked i.e. fooled into incorrectly giving high reward; adding a debate opponent reduces this. Many of the most impressive capabilities of current AI systems are produced by training on crisp tasks, like math and coding, where task success can be automatically verified. However, much of AI behavior that we actually care about is in some sense fuzzy , even for the most classical crisp tasks. For example, a coding agent should produce maintainable code, not just code that passes tests. More crucially, a coding agent should not learn to pass tests at all costs, especially by subverting the original intent of the user. More
Paper: Debate Training Reduces Reward Hacking in RLAIF Linkpost for GDM Alignment blogpost Work done by the GDM Amplified Oversight team ( we're hiring ). TL;DR : When you RL against an LLM judge, the judge gets hacked i.e. fooled into incorrectly giving high reward; adding a debate opponent reduces this. Many of the most impressive capabilities of current AI systems are produced by training on crisp tasks, like math and coding, where task success can be automatically verified. However, much of AI behavior that we actually care about is in some sense fuzzy , even for the most classical crisp tasks. For example, a coding agent should produce maintainable code, not just code that passes tests. More crucially, a coding agent should not learn to pass tests at all costs, especially by subverting the original intent of the user.
Read paper →
arXiv Aug 19, 2026

When Safety Overrides Vision: Exploring Dynamics between Vision Influence and Safety Alignment in Vision-Language Models

Mehak Gupta, Tanmoy Chakraborty

Aligned vision-language models (VLMs) are designed to balance grounded visual reasoning with safe generation behavior. However, we observe a striking phenomenon: under safety-constrained instruction, models frequently abstain from answering questions that remain correctly answerable under default instruction despite receiving identical image-question inputs. This raises a fundamental question: does safety alignment suppress perceptual grounding itself, or does visual evidence remain internally available while generation is redirected toward abstention? In this work, we investigate the internal decoding dynamics underlying safety-induced abstention in aligned VLMs. Across multiple architectures and multimodal benchmarks, we show that abstained generations remain consistently influenced by visual evidence throughout decoding, indicating that perceptual grounding is largely preserved despite refusal behavior. We further demonstrate that, although the representational organization of refusal differs substantially across architectures, safety-constrained instruction consistently alters late-stage hidden-state dynamics toward refusal-oriented decoding. More
Aligned vision-language models (VLMs) are designed to balance grounded visual reasoning with safe generation behavior. However, we observe a striking phenomenon: under safety-constrained instruction, models frequently abstain from answering questions that remain correctly answerable under default instruction despite receiving identical image-question inputs. This raises a fundamental question: does safety alignment suppress perceptual grounding itself, or does visual evidence remain internally available while generation is redirected toward abstention? In this work, we investigate the internal decoding dynamics underlying safety-induced abstention in aligned VLMs. Across multiple architectures and multimodal benchmarks, we show that abstained generations remain consistently influenced by visual evidence throughout decoding, indicating that perceptual grounding is largely preserved despite refusal behavior. We further demonstrate that, although the representational organization of refusal differs substantially across architectures, safety-constrained instruction consistently alters late-stage hidden-state dynamics toward refusal-oriented decoding.
Read paper →
arXiv Aug 19, 2026

Mechanistic Interpretability of Structure-Aware Numerical Reasoning in LLaMA 3.1 8B

Rahul Chowdhury, Timothy A Rupprecht, Senhao Cao, Jiahao Liu, Octavia Camps, David Bau, Pu Zhao, Yanzhi Wang

Recent work has shown that large language models (LLMs) exhibit strong numerical sequence modeling capabilities and show promise in time-series prediction. While LLMs display in-context learning capabilities, the mechanisms with which they accomplish time-series prediction remain unclear. Specifically, whether they truly understand the underlying structure, which at a minimum requires reasoning over first differences in the sequence of numbers. To study this, we investigate Llama 3.1-8B from a mechanistic interpretability point of view. Mechanistic interpretability is an emerging field concerned with the reverse engineering of the algorithms learned by neural networks such as LLMs. To assess Llamas' numerical sequence modeling capabilities and to facilitate our mechanistic interpretability analysis, we create a sequence modeling task that cannot be solved without picking up structural cues. Specifically, we sample n random numbers and repeat them with an offset. More
Recent work has shown that large language models (LLMs) exhibit strong numerical sequence modeling capabilities and show promise in time-series prediction. While LLMs display in-context learning capabilities, the mechanisms with which they accomplish time-series prediction remain unclear. Specifically, whether they truly understand the underlying structure, which at a minimum requires reasoning over first differences in the sequence of numbers. To study this, we investigate Llama 3.1-8B from a mechanistic interpretability point of view. Mechanistic interpretability is an emerging field concerned with the reverse engineering of the algorithms learned by neural networks such as LLMs. To assess Llamas' numerical sequence modeling capabilities and to facilitate our mechanistic interpretability analysis, we create a sequence modeling task that cannot be solved without picking up structural cues. Specifically, we sample n random numbers and repeat them with an offset.
Read paper →
arXiv Aug 18, 2026

Language Has Two Parameters: Narrative-Induced Semantic Plasticity and Phase-Sensitive Interpretation

Hollis Robbins

Reading fiction or encountering narrative generally does not merely add information. The encounter changes the reader. This paper proposes that encounters alter persistent relations among simultaneously active meanings, producing individual and shared histories that population-trained language models do not necessarily retain. A model may be told of an encounter and reproduce its consequences while the history remains in context; this is not the same as being changed by the encounter. This paper formalizes this missing relational state as phase, sets out testable predictions about encounter order, quotation, and suppressed meanings, and argues that future AI agents will need persistent semantic states indexed to particular individuals and relationships. The matching risk is semantic poisoning: an attack that re-signs relations among meanings already present. More
Reading fiction or encountering narrative generally does not merely add information. The encounter changes the reader. This paper proposes that encounters alter persistent relations among simultaneously active meanings, producing individual and shared histories that population-trained language models do not necessarily retain. A model may be told of an encounter and reproduce its consequences while the history remains in context; this is not the same as being changed by the encounter. This paper formalizes this missing relational state as phase, sets out testable predictions about encounter order, quotation, and suppressed meanings, and argues that future AI agents will need persistent semantic states indexed to particular individuals and relationships. The matching risk is semantic poisoning: an attack that re-signs relations among meanings already present.
Read paper →
arXiv Aug 18, 2026

Interpretable Humans, Alien LLMs: Expert Analysis of Latent Structures in Assessment Responses

Alona Strugatski, Licol Zeinfeld, Jason Cooper, Shelley Rap, Gil Schwarts, Giora Alexandron

The evaluation of large language models (LLMs) relies heavily on human-designed assessments, implicitly assuming that AI and humans employ similar underlying cognitive constructs. Challenging this assumption, we investigate whether the latent factors governing LLM performance carry the same substantive, human-interpretable meaning as the cognitive constructs governing human learners. Using responses from humans and six LLMs across quantitative reasoning and chemistry assessments, we conducted Exploratory Factor Analysis (EFA) separately for both groups. Subject-Matter Experts (SMEs) then blindly evaluated the resulting factor graphs to ascribe pedagogical meaning to the emerged constructs. SMEs successfully interpreted most of the human-derived factors. Conversely, they could not ascribe meaning to any LLM-derived factors in quantitative reasoning and interpreted only half of the LLM factors in chemistry. By combining data-driven EFA with blind expert interpretation, this framework shows that LLMs frequently operate on statistically opaque mechanisms distinct from human reasoning. More
The evaluation of large language models (LLMs) relies heavily on human-designed assessments, implicitly assuming that AI and humans employ similar underlying cognitive constructs. Challenging this assumption, we investigate whether the latent factors governing LLM performance carry the same substantive, human-interpretable meaning as the cognitive constructs governing human learners. Using responses from humans and six LLMs across quantitative reasoning and chemistry assessments, we conducted Exploratory Factor Analysis (EFA) separately for both groups. Subject-Matter Experts (SMEs) then blindly evaluated the resulting factor graphs to ascribe pedagogical meaning to the emerged constructs. SMEs successfully interpreted most of the human-derived factors. Conversely, they could not ascribe meaning to any LLM-derived factors in quantitative reasoning and interpreted only half of the LLM factors in chemistry. By combining data-driven EFA with blind expert interpretation, this framework shows that LLMs frequently operate on statistically opaque mechanisms distinct from human reasoning.
Read paper →
RAND Aug 18, 2026

RAND Outlines Layered Defense Strategy to Mitigate AI-Enabled Bioweapon Risk

Press Aug 18, 2026 No single safeguard can prevent an actor from using artificial intelligence to help build a biological weapon, but nine interventions layered together could meaningfully lower the risk of a high consequence AI-enabled biological attack. More
Press Aug 18, 2026 No single safeguard can prevent an actor from using artificial intelligence to help build a biological weapon, but nine interventions layered together could meaningfully lower the risk of a high consequence AI-enabled biological attack.
Read paper →
RAND Aug 18, 2026

Building a Defense-in-Depth Biosecurity Strategy for the AI Era

Research Aug 18, 2026 The authors develop a layered mitigation strategy for responding to the threat of misusing artificial intelligence–enabled biology. They examine mitigations’ strengths and weaknesses, describe areas of mutual reinforcement, and propose next steps. More
Research Aug 18, 2026 The authors develop a layered mitigation strategy for responding to the threat of misusing artificial intelligence–enabled biology. They examine mitigations’ strengths and weaknesses, describe areas of mutual reinforcement, and propose next steps.
Read paper →
RAND Aug 17, 2026

The State of Public Education in 2026 in Five Charts

Commentary Aug 17, 2026 RAND has spent more than a decade surveying educators across the United States and has expanded this effort to include the perspectives of students and parents. Among the hundreds of survey questions we fielded this past year, the results depicted in these five charts stood out the most. More
Commentary Aug 17, 2026 RAND has spent more than a decade surveying educators across the United States and has expanded this effort to include the perspectives of students and parents. Among the hundreds of survey questions we fielded this past year, the results depicted in these five charts stood out the most.
Read paper →
RAND Aug 17, 2026

AI Development in the United States and China: Evidence from a New Dataset of AI Developer Firms

Research Aug 17, 2026 The authors sought to systematically characterize commercial artificial intelligence ecosystems within the United States and China and thus broaden the empirical foundation for business intelligence and policy analysis. More
Research Aug 17, 2026 The authors sought to systematically characterize commercial artificial intelligence ecosystems within the United States and China and thus broaden the empirical foundation for business intelligence and policy analysis.
Read paper →