Balancing helpfulness and safety (harmlessness) is a critical challenge in aligning large language models (LLMs). Current approaches often decouple these two objectives, training separate preference models for helpfulness and safety, while framing safety as a constraint within a constrained Markov Decision Process (CMDP) framework. This paper identifies a potential issue when using the widely adopted expected safety constraints for LLM safety alignment, termed "safety compensation", where the constraints are satisfied on expectation, but individual prompts may trade off safety, resulting in some responses being overly restrictive while others remain unsafe. To address this issue, we propose Rectified Policy Optimization (RePO), which replaces the expected safety constraint with critical safety constraints imposed on every prompt. At the core of RePO is a policy update mechanism driven by rectified policy gradients, which penalizes the strict safety violation of every prompt, thereby enhancing safety across nearly all prompts. Our experiments demonstrate that RePO outperforms strong baseline methods and significantly enhances LLM safety alignment.

Despite the growing integration of retrieval-enabled AI agents into society, their safety and ethical behavior remain inadequately understood. In particular, the growing integration of LLMs and AI agents with external information sources and real-world environments raises critical questions about how they engage with and are influenced by these external data sources and interactive contexts. This study investigates how expanding retrieval access--from no external sources to Wikipedia-based retrieval and open web search--affects model reliability, bias propagation, and harmful content generation. Through extensive benchmarking of censored and uncensored LLMs and AIAgents, our findings reveal a consistent degradation in refusal rates, bias sensitivity, and harmfulness safeguards as models gain broader access to external sources, culminating in a phenomenon we term safety degradation. Notably, retrieval-enabled agents built on aligned LLMs often behave more unsafely than uncensored models without retrieval. This effect persists even under strong retrieval accuracy and prompt-based mitigation, suggesting that the mere presence of retrieved content reshapes model behavior in structurally unsafe ways. These findings underscore the need for robust mitigation strategies to ensure fairness and reliability in retrieval-enabled and increasingly autonomous AI systems. Content Warning: This paper contains examples of harmful language.

To recognize and mitigate the harms of generative AI systems, it is crucial to consider whether and how different societal groups are represented by these systems. A critical gap emerges when naively measuring or improving who is represented, as this does not consider how people are represented. In this work, we develop GAS(P), an evaluation methodology for surfacing distribution-level group representational biases in generated text, tackling the setting where groups are unprompted (i.e., groups are not specified in the input to generative systems). We apply this novel methodology to investigate gendered representations in occupations across state-of-the-art large language models. We show that, even though the gender distribution when models are prompted to generate biographies leads to a large representation of women, even representational biases persist in how different genders are represented. Our evaluation methodology reveals that there are statistically significant distribution-level differences in the word choice used to describe biographies and personas of different genders across occupations, and we show that many of these differences are associated with representational harms and stereotypes. Our empirical findings caution that naively increasing (unprompted) representation may inadvertently proliferate representational biases, and our proposed evaluation methodology enables systematic and rigorous measurement of the problem.

Hebbian learning is a biological principle that intuitively describes how neurons adapt their connections through repeated stimuli. However, when applied to machine learning, it suffers serious issues due to the unconstrained updates of the connections and the lack of accounting for feedback mediation. Such shortcomings limit its effective scaling to complex network architectures and tasks. To this end, here we introduce the Structural Projection Hebbian Representation (SPHeRe), a novel unsupervised learning method that integrates orthogonality and structural information preservation through a local auxiliary nonlinear block. The loss for structural information preservation backpropagates to the input through an auxiliary lightweight projection that conceptually serves as feedback mediation while the orthogonality constraints account for the boundedness of updating magnitude. Extensive experimental results show that SPHeRe achieves SOTA performance among unsupervised synaptic plasticity approaches on standard image classification benchmarks, including CIFAR-10, CIFAR-100, and Tiny-ImageNet. Furthermore, the method exhibits strong effectiveness in continual learning and transfer learning scenarios, and image reconstruction tasks show the robustness and generalizability of the extracted features. This work demonstrates the competitiveness and potential of Hebbian unsupervised learning rules within modern deep learning frameworks, demonstrating the possibility of efficient and biologically inspired learning algorithms without the strong dependence on strict backpropagation.

The remarkable progress in neural-network-driven visual data generation, especially with neural rendering techniques like Neural Radiance Fields and 3DGaussian splatting, offers a powerful alternative to GANs and diffusion models. These methods can generate high-fidelity images and lifelike avatars, highlighting the need for robust detection methods. However, the lack of any large dataset containing images from neural rendering methods becomes a bottleneck for the detection of such sophisticated fake images. To address this limitation, we introduce NeuroRenderedFake, a comprehensive benchmark for evaluating emerging fake image detection methods. Our key contributions are threefold: (1) A large-scale dataset of fake images synthesized using state-of-the-art neural rendering techniques, significantly expanding the scope of fake image detection beyond generative models; (2) A cross-domain evaluation protocol designed to assess the domain gap and common artifacts between generative and neural rendering-based fake images; and (3) An in-depth spectral energy analysis that reveals how frequency domain characteristics influence the performance of fake image detectors. We train representative detectors, based on spatial, spectral, and multimodal architectures, on fake images generated by both generative and neural rendering models. We evaluate these detectors on 15 groups of fake images synthesized by cutting-edge neural rendering models, generative models, and combined methods that can exhibit artifacts from both domains. Additionally, we provide insightful findings through detailed experiments on degraded fake image detection and the impact of spectral features, aiming to advance research in this critical area.

For Iran's Athletes, There Is No Separating Sports From Politics From defections and protests to moments of national pride, the 2026 World Cup arrives amid decades of tension between identity and the state. Iran's national soccer team has made its 2026 World Cup debut amid a tumultuous backdrop: an abrupt and tentative ceasefire after months of war, an extraordinary set-up in Mexico after the US prevented the team from residing in-country between matches, and political uncertainty that has now expanded to the international stage. But for many Iranians, professional sports have always sat at an intersection between athleticism, identity, and politics. From sporting defections and political activism to moments of immense national sporting pride, the trajectory of Iranian sports underscores what's at stake this World Cup. The Iranian team, on Tuesday morning, drew 2-2 in their debut against New Zealand and will next face Belgium and Egypt, traveling to and from Mexico in between.

WIRED goes on location--and on camera--with the cult hit. On March 16, 2026, at 5:45 pm in a leafy suburb of Atlanta called Sandy Springs, police pound on the door of a neglected French Country-style mansion, rifles at the ready, bodycams rolling. Minutes earlier, a distress call came from someone claiming to be hiding from a gunman in the mansion's downstairs bathroom. The dispatcher heard a gunshot ring out in the distance, then the line disconnected. "Open the door!" an officer yells. A calm young man with a mullet and woolly eyebrows steps out, hands raised. The police ask him who else is in the house. "Just my friends," he replies, as seven other young people, men and women, silently file out behind him, less evidently relaxed. They remain outside while two officers search the house. Inside the mansion there are no immediate signs of a massacre, but the decor alone arouses suspicion. All of the windows are frosted over, so only a chilly light leaks in. The place is a mess, and the walls are adorned with lurid, seemingly AI-generated art: a frowning baby holding an assault rifle, a rubber ducky bobbing in a mug of what looks like black coffee, a lidless and levitating eyeball crying into a martini glass. The rooms are painted primary colors, grass green and cherry red, like a kindergarten class. A vape dangles from a doorframe by a chain, suspended at mouth level. The pantry is practically empty. The bedroom is a dormitory featuring seven identical twin beds. No one is hiding in the bathroom. The call, it seems, was a prank. The police return to the driveway and ask, "What is it that you guys are doing here?" "We're just livestreaming," says a man in a camo hat named Matt. "You guys don't have any firearms or anything inside the house?" There are guns in the house, Matt says, for self-defense. Fans of their livestream can be obsessive, he explains, and tend to have perverse ideas about jokes. The officer asks to see their weapons, and they go downstairs. The room is cluttered with ergonomic swivel chairs, desks strewn with takeout containers and energy drinks, two flatscreen TVs, and a dozen computer monitors.

Direct preference optimization (DPO) is widely used as a simple and stable method for aligning large language models (LLMs) with human preferences. This paper investigates a generalized DPO loss that enables a policy model to match the target policy from a likelihood ratio estimation perspective. The ratio of the target policy provides a unique identification of the policy distribution without relying on reward models or partition functions. This allows the generalized loss to retain both simplicity and theoretical guarantees, which prior work such as f-PO fails to achieve simultaneously. We propose Bregman preference optimization (BPO), a generalized framework for ratio matching that provides a family of objective functions achieving target policy optimality.

The DeceptionBench is designed as a research benchmark to systematically study deception behaviors in LLMs, fostering a deeper understanding of their decision-making processes in real-world scenarios. Our primary intent is to provide a standardized, transparent tool for the research community to evaluate and improve LLMs' ethical alignment, not to enable or encourage deceptive practices. To prevent potential misuse by malicious actors, we commit to publicly releasing all evaluation data under an open license. This transparency ensures that DeceptionBench's methodology and outcomes are subject to scrutiny, replication, and improvement by the research community, reducing the risk of hidden exploitation. By prioritizing openness, we aim to advance responsible AI development while safeguarding against misuse in harmful contexts. The field of Large Language Models (LLMs) has undergone remarkable evolution in recent years, reshaping the landscape of natural language processing.

Despite the remarkable advances of Large Language Models (LLMs) across diverse cognitive tasks, the rapid enhancement of these capabilities also introduces emergent deception behaviors that may induce severe risks in high-stakes deployments. More critically, the characterization of deception across realistic real-world scenarios remains underexplored. To bridge this gap, we establish DeceptionBench, the first benchmark that systematically evaluates how deceptive tendencies manifest across different societal domains, what their intrinsic behavioral patterns are, and how extrinsic factors affect them. Specifically, on the static count, the benchmark encompasses 150 meticulously designed scenarios in five domains, i.e., Economy, Healthcare, Education, Social Interaction, and Entertainment, with over 1,000 samples, providing sufficient empirical foundations for deception analysis. On the intrinsic dimension, we explore whether models exhibit self-interested egoistic tendencies or sycophantic behaviors that prioritize user appeasement. On the extrinsic dimension, we investigate how contextual factors modulate deceptive outputs under neutral conditions, reward-based incentivization, and coercive pressures. Moreover, we incorporate sustained multi-turn interaction loops to construct a more realistic simulation of real-world feedback dynamics. Extensive experiments across LLMs and Large Reasoning Models (LRMs) reveal critical vulnerabilities, particularly amplified deception under reinforcement dynamics, demonstrating that current models lack robust resistance to manipulative contextual cues and the urgent need for advanced safeguards against various deception behaviors.