Safety & alignment

arXiv:2607.26067v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly used for estimating item difficulty in educational assessment. However, it remains unclear whether such estimates reflect how learners actually experience difficulty. This study investigates the alignment between LLM-generated difficulty ratings and empirical student performance on basic mathematics tasks. Four widely used LLM-based systems generated difficulty ratings on a 1-100 scale for 32 arithmetic

arXiv:2607.26654v1 Announce Type: cross Abstract: Post-training alignment is often shallow, eroding under fine-tuning. Whether midtraining interventions, cleanly isolated from post-training, can produce durable alignment remains untested. We test this via constitutional midtraining: inserting principled, values-based content into midtraining against a replay-only control at 120B scale. Our 394M-token constitutional corpus, built from Anthropic's Constitution, uses a 2x2 factorial design (curricu

arXiv:2604.24155v4 Announce Type: replace Abstract: The project of aligning machine behavior with human values raises a basic problem: whose moral expectations should guide AI decision-making? Much alignment research assumes that the appropriate benchmark is how humans themselves would act in a given situation. Studies of agent-type value forks challenge this assumption by showing that people do not always judge humans and AI systems identically. This paper extends that challenge by examining tw

arXiv:2507.04398v3 Announce Type: replace-cross Abstract: Generative AI is becoming part of academic writing, but its educational value depends on how control is shared between learner and system. This study examined an agency gap: performance differences that may arise when AI agent initiative is misaligned with learners' generative AI literacy. Seventy-nine medical and nursing students completed two multimodal analytical writing tasks using healthcare simulation data visualisations. They were

Weng previously served as the VP of AI Safety Research at OpenAI.

The agentic AI playing field was heavily tilted toward offense, so researchers began using red team agents to help teach their blue counterparts.

I watched a new tool try to get around the model safeguards of four major frontier companies. You might be surprised by how they performed.

Scientific images are the core elements of presenting experimental conclusions, elaborating system architecture, and supporting comparative arguments in scientific papers. However, existing image quality assessment (IQA) methods are predominantly designed for natural photographs or AI-generated content, which cannot be directly applied to scientific papers. The few existing studies on scholarly charts remain confined to visual-surface comparisons, failing to verify caption alignment, citation re

Scientific figure assessment in peer review differs fundamentally from general image quality evaluation: a figure must be visually legible, faithfully support the manuscript's claims, and communicate evidence with a clear visual hierarchy. However, if we apply traditional image assessment methods to scientific figure quality assessment, limitations emerge: classic IQA models capture perceptual quality or aesthetics but cannot judge whether a figure serves the paper's scientific argument; CLIP-ba

It started as China vs. the US, but it's become a face-off between two fundamentally different ways of building LLMs. And the safety of everything is on the line.

Multimodal Large Language Models (MLLMs) rely on a projector to align visual representations with the language embedding space, making it central to cross-modal understanding. In Multimodal Continual Instruction Tuning (MCIT), however, shifting visual distributions and evolving instruction semantics cause this shared projector to drift, leading to projector-level forgetting, an issue largely overlooked by methods that focus primarily on the LLM backbone. We introduce Progressive Multimodal Align

Federated Learning (FL) is vulnerable to backdoor attacks because of its distributed nature in edge computing scenarios. Existing defense methods show limited efficacy as they overlook the deviations among benign local updates caused by statistical heterogeneity and the stealthiness of backdoor attacks. To tackle these issues, we propose FedDAB, a two-phase method that combines local contrastive regularization with alignment checking, to defend against backdoor attacks. In the first phase, FedDA

Federated learning (FL) enables collaborative learning over decentralized data silos without centralizing raw data. However, heterogeneous local architectures often induce non-aligned representation spaces, making it difficult to transfer global knowledge across silos. Existing paradigms share this knowledge as model parameters, distilled predictions, or class prototypes, yet all encode it in an absolute space that must be aligned across clients. Heterogeneous backbones break this alignment, so

Earlier this month, OpenAI gave several of its AI models a task: complete a test designed to measure their cybersecurity capabilities. It put the systems in a sandboxed environment without an internet connection and set them off to work. What happened next is almost laughably silly - but also, as Adam Gleave, cofounder and CEO […]

Post-training alignment is often shallow, eroding under fine-tuning. Whether midtraining interventions, cleanly isolated from post-training, can produce durable alignment remains untested. We test this via constitutional midtraining: inserting principled, values-based content into midtraining against a replay-only control at 120B scale. Our 394M-token constitutional corpus, built from Anthropic's Constitution, uses a 2x2 factorial design (curriculum ordering x deliberative reasoning) to produce

Safety classifiers ("guards") are the dominant black-box defense for vision-language models, yet they judge an input's surface form, not its meaning: a harmful request re-encoded as set theory, formal logic, a rare language, code, or an image of text slips past a guard that would block it in plain language -- the decode gap. The natural fix is a guard-agnostic recover-and-decode amplifier that transcribes image content and restates encoded text into its plain payload before the guard, so any off

Audio-capable foundation models enable end-to-end spoken interaction, but they also introduce safety risks beyond transcript content. It remains unclear how much jailbreak capability can arise from matched-text variation in speech delivery rather than from lexical rewriting or broader style transfer. We study this question by holding transcript content fixed and varying six speech-delivery presets whose acoustic attributes may co-vary. We present PJ-Break, a black-box evaluation protocol with pr

Wi-Fi-based human activity recognition has advanced substantially, but most existing methods assume a closed set of activities and require labeled Wi-Fi samples for every target class, limiting their ability to recognize unseen activities. We present Zero-Fi, a contrastive signal-language alignment framework for zero-shot Wi-Fi-based human activity recognition. Zero-Fi learns unified representations from complementary Wi-Fi signal features and aligns them with the semantic representations of nat

The Food and Drug Administration's (FDA) plans to centralize and enhance key functions across the agency. These changes will reduce redundancies, improve efficiency, and advance alignment to better serve the American public.

Researchers propose focusing on identification of certain cognitive elements in LLMs that indicate when AI systems may take an unwanted action.

The Chinese AI startup’s massive new model is challenging OpenAI and Anthropic, fueling a debate over AI safety.

The Chinese AI startup’s massive new model is challenging OpenAI and Anthropic, fueling a debate over AI safety.

Alignment training, model organisms, and toy models are usually treated as separate research areas. But projects in all three frequently use supervised fine-tuning (SFT) to pursue the same underlying goals. When projects share a goal, we should test whether lessons learned from one area transfer to the other areas. We study three such transfers, each taking a lesson developed in one SFT setting and testing it in another. First, we port a lesson about behavior generalization from alignment traini

Automated molecular structure elucidation from infrared (IR) spectroscopy data has seen significant advancements in recent years, but its broad applicability is limited by a reliance on pre-determined chemical formulas provided as auxiliary model inputs. This limits model predictions to isomer identification rather than full molecular structure prediction. Although transformer models have been shown to identify molecular isomers with high accuracy, their reliability for unconstrained structure e

Vision-Language-Action (VLA) models have shown strong potential for general robot manipulation, but most existing models rely on 2D visual-language backbones and lack fine-grained 3D understanding of target objects, especially under occlusion, pose variation, scale changes, and precise spatial interaction. We propose an object-centric 3D representation alignment framework built upon $π_0$, using SAM3D as a frozen 3D teacher to provide target-object 3D priors during training. Specifically, we loc

We introduce \textbf{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

The industry's response to the OpenAI incident reveals a growing divide over the future of AI safety.

We propose Rashomon Alignment (RA), a new measure to assess functional similarity between two models. Existing functional similarity measures are distributional, quantifying differences between outputs of models applied to real-world data. However, these measures can be regarded as ecologically valid only for regions in the input space represented by the available data. We introduce a geometrical perspective on functional model similarity, which estimates it across the entire data space, offerin

Public services face growing pressure to adopt artificial intelligence (AI) to close the gap between rising demand and falling resources. That pressure has intensified with general-purpose AI (GPAI): AI built on large language models that can be directed by prompt alone to perform an effectively unbounded range of tasks. We argue that the properties that make these models attractive - their generality, accessibility, and low deployment cost - undermine the conditions under which AI safety has hi

Decoding speech information directly from scalp electroencephalography (EEG) into text provides a potential non-invasive neural communication pathway for individuals with severe speech and motor impairments. Compared with invasive approaches such as electrocorticography, EEG is safer and more widely deployable, yet substantially more challenging to decode.This challenge is exacerbated for Chinese sentence decoding, which must handle a high-dimensional output space with thousands of characters, s

Adapting Large Language Models (LLMs) to specialized domains often incurs an alignment tax, as fine-tuning on domain-specific tasks can cause catastrophic forgetting and substantially degrade performance on general tasks. We propose MemSFT, which mitigates the alignment tax by decoupling domain specialization from backbone parameter updates through a plug-and-play parametric memory. The memory is trained to imitate the behavior of a non-parametric retriever operating over domain data, thereby me

Perception tasks for autonomous vehicles need to work satisfactorily in adverse weather conditions. Due to lack of real-world weather datasets, weather simulations are a promising alternative. To ensure simulations closely mirror real-world weather data, it's crucial that they represent the same weather characteristics, including severity and particle positioning, across different sensors. To achieve this, we propose the Reference Dataset Alignment Method (ReDAM) for weather intensity alignment

Entity alignment (EA) identifies entities across knowledge graphs (KGs) that refer to the same real-world object. Conventional EA methods mainly exploit explicit graph structures and textual fields, which often provide insufficient semantic understanding to recognize the same entity under heterogeneous descriptions and distinguish it from semantically similar entities. Although large language models (LLMs) offer deeper entity understanding, existing LLM-based EA methods largely use this capabili

Multimodal drug discovery enables drug representation learning beyond chemical structure by incorporating cellular responses such as gene expression and cell morphology. However, direct fusion and instance-level contrastive alignment may mix mechanism-related signals with modality-specific noise and incorrectly separate structurally dissimilar but biologically related compounds. This limitation can obscure transferable mechanism patterns required for predicting the properties of unseen compounds

This paper presents ContractHIL-HLS, a contract-aligned multi-agent workflow for practical high-level synthesis (HLS) engineering. The workflow makes three contributions. First, it introduces a structured contract as the semantic-alignment and task-execution artifact that translates natural language requirements into explicit interfaces, constraints, validation checks, and rollback rules. Second, it incorporates hardware information into the feedback loop by feeding HLS, Vivado, PYNQ runtime, po

arXiv:2607.23519v1 Announce Type: new Abstract: 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-fir

arXiv:2607.24391v1 Announce Type: new Abstract: 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 sociolo

arXiv:2607.22953v1 Announce Type: cross Abstract: Modern AI systems bring societal risks such as mass surveillance, extreme concentrations of power, and loss of user autonomy---calling into question a model where third-parties collect and control massive amounts of user data. Users require a sovereign system to securely own, govern, and disclose their context while remaining compliant across regulated domains with strict provenance, interpretability, and policy adherence. Perspective-aware AI ap

Generative recommendation commonly represents items using fixed-length semantic identifiers (SIDs) constructed through clustering and quantization. However, these artificial codes may overcompress item semantics, remain misaligned with pretrained LLM vocabularies, and require costly autoregressive decoding. In light of this, we propose VaLiDRec, a generative recommendation framework based on variable-length, LLM-aligned semantic identifiers. VaLiDRec constructs SIDs directly from informative nat