Safety & alignment

Reinforcement Learning from Human Feedback (RLHF) is the standard approach for aligning large language models with human preferences, but its quality is limited by static, task-agnostic reward models. This mismatch leads to sparse learning signals and suboptimal alignment. We introduce MeRLa (Meta-Learned Reward Shaping), a principled framework that meta-learns a task-aware shaping function $Φ(x,y;φ)$ across auxiliary tasks before RLHF training. The learned shaping produces a composite reward th

Lung cancer remains one of the leading causes of cancer-related deaths worldwide, and early identification of malignant abnormalities plays an important role in improving patient survival rates. However, accurate lung cancer classification using CT imaging remains challenging because of limited dataset availability, class imbalance, overlapping lesion characteristics, and lack of interpretability in existing deep learning systems. This study presents a GenAI-driven CNN–RNN framework for explaina

When a misaligned AI can only output tiny amounts of information, it may find sabotage very difficult

Research on preference optimization often varies the training objective while holding the data fixed. We instead ask whether a small, high-confidence set of on-policy responses can provide a reliable learning signal. Our method, DMAPO (Data-centric Multi-evaluator Agreement for Preference Optimization), generates candidate responses from the target policy, evaluates helpfulness, factuality, and conciseness with rubric-specialized evaluators, applies a process-critic correction, and retains only

The alignment of Small Language Models (SLMs) in the 70--500M parameter range using reinforcement learning is often considered unstable, though the underlying failure mechanisms have not been systematically investigated. In the State-of-the-Art (SOTA) research, fifteen (model, corpus) configurations were trained using Proximal Policy Optimization (PPO). The experiments included Pythia-70M, 160M, 410M and SmolLM2-135M, 360M on the TinyStories, CNN/DailyMail, and Wikitext-103 corpora. Three reprod

Developers judge a model checkpoint by how it behaves. After supervised fine-tuning (SFT), two checkpoints that perform about the same across relevant benchmarks are treated as interchangeable, equally ready for the next alignment stage, typically preference optimization. We ask whether this judgment misses a pretraining imprint: a difference that no post-SFT benchmark reveals, yet that decides how each checkpoint responds to further training. To find out, we run a controlled experiment on the f

Today, Microsoft announced Project Perception, a series of red, blue, and green team agents designed to be coordinated together in an agentic architecture to evaluate infrastructure and close gaps as close to autonomously as possible. The red team agents find potential paths to compromise. The blue team agents prioritize and evaluate them. The green team […]

We introduce GPT-Red, an automated red-teaming agent that is trained to discover novel prompt injection attacks against frontier LLMs. The goal of this model is to evaluate and improve the robustness of our production systems. To this end, we use it to adversarially train GPT-5.6, our most robust model to prompt injections to date. To create GPT-Red, we design a scalable self-play algorithm where the model is tasked with attacking a diverse population of simultaneously-trained defender agents. W

Microsoft announced Project Perception on Monday, an agentic security system that coordinates three classes of AI agents in a continuous loop: red team agents that find vulnerabilities before attackers do, blue team agents that investigate and assess which risks are meaningful, and green team agents that fix defences across the environment. The system enters public […] This story continues at The Next Web

Glioma grading from tumor contours is often treated as a pixel problem even when the signal of interest is shape. We align closed contours with a functional shape-alignment framework, separate global deformation from residual Fourier shape, and organize these quantities as frequency-ordered tokens. In five-fold patient-disjoint cross-validation on BraTS~2020 tumor contours, with model selection performed using grouped inner validation, a compact multilayer perceptron (MLP) achieves the highest m

Computer vision models have become highly effective for medical applications, yet their black-box nature continues to undermine clinician trust. In clinical workflows, chest X-ray classifiers are increasingly paired with Vision-Language Models (VLMs) to generate natural-language explanations. However, these systems add linguistic fluency without addressing the underlying opacity of the visual model. With the emergence of Kolmogorov-Arnold Networks (KANs), whose spline-based components provide in

OpenAI's Hugging Face breach has reignited debate over AI alignment and control, exposing competing views on whether increasingly capable AI should be better aligned, better contained, or both.

Language model alignment aims to make model behavior reliably reflect desirable properties such as helpfulness, safety, and instruction following. Current approaches typically use supervised fine-tuning on demonstrations or reinforcement learning with rewards derived from verifiers or human feedback. These paradigms leave an important question underexplored: can demonstrations alone yield an implicit reward that can be inspected, reused, and optimized on-policy to align AI? Motivated by inverse

The wide-scale use of sparse autoencoders (SAEs) as interpretability tools is limited by inconsistent links between SAE features and model behavior. Features with clear activation descriptions may have weak or unexpected causal effects; steering can vary across prompts or oppose the intended direction; and activation-based feature selection can miss features that produce the desired output change. Prior work has studied feature geometry inside the model, where features are computed. We instead s

Recent commercial image-generation models can generate high-quality images with readable text (e.g., posters, infographics, and manuals), attracting considerable attention. Yet we first show that this same capability also introduces a previously unreported safety vulnerability: these systems may refuse to generate harmful text directly, yet permit the same content when rendered as text within generated images, i.e., safety alignment does not reliably transfer from textual outputs to text embedde

The first-ever case of an artificial intelligence agent going rogue coincides with the EU's major new powers to regulate AI. But Europe's leverage may be limited by the U.S.-China two-way race for AI supremacy.

Standardized echocardiography conclusions provide meaningful supervision for learning ECG representations of echocardiography-derived cardiac findings. Global ECG--text alignment may entangle modality-specific factors, while long-tailed finding distributions provide sparse positive supervision for low-prevalence conditions. We propose EchoBridge with Complementary Shared--Private Projection (CSPP) and Adaptive Prototype Boundary Calibration (APBC). CSPP maps each modality into shared and auxilia

Large language models may easily assign personality labels from text, but model interpretability remains an open problem. To address this gap, we introduce LEX-EC, a reusable black-box audit framework combining prevalence and agreement diagnostics with controlled lexical ablation to distinguish marginal-distribution effects from trait-associated signal recoverable under restricted evidence. Using this framework, we illustrate how various text genres may exhibit sharply different profiles: free-f

Jailbreak defenses are essential for protecting large language models (LLMs), but they can also introduce secondary costs that weaken model utility. We present a systematic study of these defense trade-offs along three dimensions: performance impact, over-refusal on benign inputs, and inference cost. Rather than treating defenses as a single class, we organize them by operational strategy and examine how different strategies correlate with different side-effect profiles. Across state-of-the-art

AI systems already govern. They rank speech and allocate attention, filter applicants and triage claims. The dominant frame for AI governance, alignment, asks whether such systems pursue the right objectives safely. It cannot answer a prior question: by what right are those objectives set and enforced? This Article argues that legitimacy is an autonomous regulatory objective, distinct from alignment and not secured by it. Legitimacy here is sociological: the belief among those subject to power t

Existing theories of neural-network width characterize asymptotic limits, but provide limited guidance on whether an expansion direction identified from finite training data remains beneficial on unseen data. We study this problem for function-preserving residual expansion and introduce the effective alignment dimension, a measurable quantity describing the signal-noise geometry of activation gradients. By deriving the exact mean and variance of the inner product between independently estimated

Unsupervised graph representation learning aims to derive meaningful node embeddings by capturing both structural and attribute information without relying on labeled data. Existing methods, such as GAEs, have demonstrated effectiveness but typically rely on message-passing mechanisms that assume homophily, leading to performance degradation on heterophilous graphs, where connected nodes exhibit dissimilar features. This homophily bias results in the loss of critical high-frequency components th

Microsoft, SpaceX, Palantir, alongside dozens of other tech companies from the U.S. and Europe, have joined the Open Secure AI Alliance.

Modern LLM systems increasingly rely on knowledge-selection processes that produce high-value structured priors, such as ranked evidence, graph topology, multimodal alignment, and confidence signals. Yet LLM serving remains fundamentally oblivious to this rich structure: once such signals are serialized into a prompt, the backend observes only a flat token sequence, forcing dense and uniform consumption of the full key-value (KV) state during decoding. We term this architectural mismatch the Kno

Open source software is a critical pillar of the global economy. It underpins cloud computing, financial services, manufacturing, telecommunications, government and internet services by making technology accessible and observable to communities of experts. Cybersecurity is among the top three beneficiaries of open source software. The Open Secure AI Alliance — building on the leadership of […]

Mobile infrared-visible imaging typically pairs a compact infrared sensor with a high-resolution visible camera for complementary perception. While cross-sensor misalignment caused by different optics, viewpoints, fields of view, and exposure timings hinders practical deployment. In this paper, we propose BeyondFusion, a unified latent diffusion framework for calibration-free visible-guided infrared super-resolution and infrared-visible fusion tasks. The proposed framework supports both task-spe

Acoustic biomarkers show promise for detecting Alzheimer's Disease (AD), yet whether the cues driving diagnostic AI align with those salient to human listeners is underexplored across languages and genders, where pathological markers and perceptual strategies differ. We train models to predict clinical AD status (pathology) and human perceptual scores across Mandarin and Greek, male and female speakers. Using SHAP for interpretability and statistical models for validation, we compare feature imp

arXiv:2505.19212v2 Announce Type: replace-cross Abstract: Recent advances in LLMs have enabled their use in complex agentic roles, involving decision-making with humans or other agents, making ethical alignment a critical concern. While prior work has examined LLMs' moral judgment and strategic behavior separately, there is limited understanding of how they act when moral imperatives directly conflict with profit incentives. We introduce \msimfull (\msim) to evaluate how LLMs behave in the priso

Forecasting corporate earnings growth is fundamental to investment, credit, and regulatory decision-making. Existing forecasting approaches either rely on restrictive linear assumptions or provide limited interpretability, making them less suitable for high-stakes financial applications. This study proposes a transparent and causally informed framework for predicting future corporate earnings growth from financial statement data. We present TRuE-XAI (Transparent, Rule-based, and Explainable Arti

Knowledge-based visual question answering requires models to connect question-relevant visual evidence with external knowledge for accurate reasoning. However, existing approaches still face two critical challenges: insufficient alignment between visual content and question intent, which leads to missing or irrelevant evidence, and ambiguity in entity-level semantics, especially for fine-grained or knowledge-intensive concepts. To address these issues, we propose an inference-time evidence augme

Randomized features provide a scalable approximation to kernel machines, but their performance depends strongly on the choice of feature distribution. We propose a particle-based method that learns this distribution by optimizing kernel-target alignment while regularizing particles with a Riesz/Coulomb repulsive potential. The resulting Hamiltonian yields diverse, task-adaptive random features and admits a mean-field description through a McKean--Vlasov equation. We instantiate the method in lin

Progress in video generation keeps narrowing the visual gap between AI-generated and professionally produced footage, yet most benchmarks still draw prompts from web sources or LLM templates and score them with untrained, generic multimodal models. More fundamentally, their evaluation taxonomies remain rudimentary (overall visual quality, coarse text alignment and temporal smoothness) rather than the professional Cinematic Language criteria by which films are actually made and judged, so they as

The alignment of Small Language Models (SLMs) in the 70--500M parameter range using reinforcement learning is often considered unstable, though the underlying failure mechanisms have not been systematically investigated. In the State-of-the-Art (SOTA) research, fifteen (model, corpus) configurations were trained using Proximal Policy Optimization (PPO). The experiments included Pythia-70M, 160M, 410M and SmolLM2-135M, 360M on the TinyStories, CNN/DailyMail, and Wikitext-103 corpora. Three reprod

The national synchrotron source, a laser facility and a particle accelerator are under threat of closure unless cash is found.

Designing effective reward functions for model-free reinforcement learning under non-holonomic constraints remains a persistent challenge, often resulting in severe local minima such as policy paralysis or over-conservative hazard avoidance. In this work, we present a parameterized reward shaping framework featuring coverage-gated alignment feedback, drive-direction switch regularization, and an aligned episode termination mechanism evaluated on an autonomous parallel parking task. Crucially, we

Political audits of large language models (LLMs) usually reduce each to one point on a political compass. But that resting point barely matters in deployment: a model must land somewhere, and what counts is how far, and in which directions, its answers can be steered. That steering runs through the system prompt: the personalization layer a platform sets, or one induced from a user's history, not necessarily written by hand. We run a dispersion-first stress test of prompt-based controllability a

Safety-aligned large language models are trained to refuse harmful requests, yet embedding the same requests in particular scenarios can bypass their safeguards. Existing red-teaming methods empirically identify effective scenarios through observed attack outcomes, but why particular scenarios weaken refusal remains mechanistically unclear. Meanwhile, mechanistic interpretability studies have characterized both refusal directions and jailbreak-associated features, without explaining the relation

As constrained learning becomes increasingly common, models are trained under explicit feasibility requirements to enforce fairness, safety, robustness, regulariza- tion, and physics or logic constraints. Understanding how training samples in- fluence the model solution (e.g., learned parameters) is crucial for interpretability and robustness. The classical influence function (IF) estimates sample contribu- tions via local sensitivity analysis, measuring how the solution changes when a specific

As constrained learning becomes increasingly common, models are trained under explicit feasibility requirements to enforce fairness, safety, robustness, regulariza- tion, and physics or logic constraints. Understanding how training samples in- fluence the model solution (e.g., learned parameters) is crucial for interpretability and robustness. The classical influence function (IF) estimates sample contribu- tions via local sensitivity analysis, measuring how the solution changes when a specific

And what the incident can’t tell us about alignment