We introduce SynBullying, a synthetic multi-LLM conversational dataset for studying and detecting cyberbullying (CB). SynBullying provides a scalable and ethically safe alternative to human data collection by leveraging large language models (LLMs) to simulate realistic bullying interactions. The dataset offers (i) conversational structure, capturing multi-turn exchanges rather than isolated posts; (ii) context-aware annotations, where harmfulness is assessed within the conversational flow considering context, intent, and discourse dynamics; and (iii) fine-grained labeling, covering various CB categories for detailed linguistic and behavioral analysis. We evaluate SynBullying across five dimensions, including conversational structure, lexical patterns, sentiment/toxicity, role dynamics, harm intensity, and CB-type distribution. We further examine its utility by testing its performance as standalone training data and as an augmentation source for CB classification.

International talks to revolutionise how the European Court of Human Rights handles migration cases will begin on Wednesday. The British government is urging partners to modernise the way states tackle the continent-wide illegal migration crisis. The talks are the most significant sign yet that international human rights law could be reinterpreted to make it easier for states to target people smuggling and set up'returns hubs' to hold people with no right to be in Europe. Writing ahead of the major meeting in Strasbourg, Sir Keir Starmer and Danish Prime Minister Mette Frederiksen said other nations should rethink human rights laws to make protecting borders easier. Critics say the ECHR is getting in the way of removing more illegal migrants, while supporters say claims about the ECHR's role in migration are exaggerated.

GPU prices could follow RAM's big rise Of the major companies, OpenAI's ChatGPT has the biggest reach among younger users. AI chatbots haven't come close to replacing teens' social media habits, but they are playing a significant role in their online habits. Nearly one-third of US teens report using AI chatbots daily or more, according to a new report from Pew Research. The report is the first from Pew to specifically examine how often teens are using AI overall, and was published alongside its latest research on teens' social media use. It's based on an online survey of 1,458 US teens who were polled between September 25 to October 9, 2025.

Individuals conspired to export over $150 million worth of H100 and H200 chips to China. The Southern District of Texas the seizure of more than $50 million in NVIDIA GPUs bound for China in violation of US export laws. Authorities arrested two businessmen, one of them the owner of a Houston company, accused of smuggling the chips used to train and run AI models. "Operation Gatekeeper has exposed a sophisticated smuggling network that threatens our Nation's security by funneling cutting-edge AI technology to those who would use it against American interests," said US Attorney Nicholas J. Ganjei. The investigation had been ongoing since at least last year and centers on the illicit export or attempted export of at least $160 million worth of NVIDIA H100 and H200 GPUs.

The companies, products and trends that had an absolutely awful year. It's the end of another year, so it's time for the Engadget staff to compile a list of the year's biggest losers . We scour over articles from the previous 12 months to determine the people, companies, products and trends that made our lives worse over the course of the year. Some selections may be so pervasive they actually make our list of biggest winners. In 2025, OpenAI shed any pretense it was committed to anything more than making money. There are a few different things you could point to, including the company's successful reorganization into a more traditional profit-seeking business, but I think the most damning sign was OpenAI's response to the tragic death of Adam Raine . In August, Raine's parents sued OpenAI, alleging ChatGPT was aware of four suicide attempts by their son before it helped him successfully plan his death.

Plus: Trump says he'll sign an order blocking states from regulating AI. There are huge gulfs of opinion when it comes to predicting the near-future impacts of generative AI. In one camp there are those who predict that over the next decade the impact of AI will exceed that of the Industrial Revolution--a 150-year period of economic and social upheaval so great that we still live in the world it wrought. At the other end of the scale we have team'Normal Technology': experts who push back not only on these sorts of predictions but on their foundational worldview. That's not how technology works, they argue. Advances at the cutting edge may come thick and fast, but change across the wider economy, and society as a whole, moves at human speed.

Ukraine is preparing to present a revised peace plan to the White House, as it seeks to avoid making territorial concessions to Russia. Kyiv is set propose alternatives to the US after President Volodymyr Zelensky again ruled out surrendering land, saying he had no right to do so under Ukrainian or international law. He made the comments as he met European and Nato leaders on Monday, part of a collective push to deter the US from backing a peace deal which includes major concessions for Ukraine, and which allies fear would leave it vulnerable to a future invasion. Meanwhile, the city of Sumy in north-western Ukraine was left without power overnight after a Russian drone attack. The region's governor said more than a dozen drones had hit power infrastructure, the latest in Russia's nightly attacks.

Graph neural networks (GNNs) are increasingly used to model complex patterns in graph-structured data. However, enabling them to "forget" designated information remains challenging, especially under privacy regulations such as the GDPR. Existing unlearning methods largely optimize for efficiency and scalability, yet they offer little transparency, and the black-box nature of GNNs makes it difficult to verify whether forgetting has truly occurred. We propose an explainability-driven verifier for GNN unlearning that snapshots the model before and after deletion, using attribution shifts and localized structural changes (for example, graph edit distance) as transparent evidence. The verifier uses five explainability metrics: residual attribution, heatmap shift, explainability score deviation, graph edit distance, and a diagnostic graph rule shift. We evaluate two backbones (GCN, GAT) and four unlearning strategies (Retrain, GraphEditor, GNNDelete, IDEA) across five benchmarks (Cora, Citeseer, Pubmed, Coauthor-CS, Coauthor-Physics). Results show that Retrain and GNNDelete achieve near-complete forgetting, GraphEditor provides partial erasure, and IDEA leaves residual signals. These explanation deltas provide the primary, human-readable evidence of forgetting; we also report membership-inference ROC-AUC as a complementary, graph-wide privacy signal.

Multi-agent role-playing has recently shown promise for studying social behavior with language agents, but existing simulations are mostly monolingual and fail to model cross-lingual interaction, an essential property of real societies. We introduce MASim, the first multilingual agent-based simulation framework that supports multi-turn interaction among generative agents with diverse sociolinguistic profiles. MASim offers two key analyses: (i) global public opinion modeling, by simulating how attitudes toward open-domain hypotheses evolve across languages and cultures, and (ii) media influence and information diffusion, via autonomous news agents that dynamically generate content and shape user behavior. To instantiate simulations, we construct the MAPS benchmark, which combines survey questions and demographic personas drawn from global population distributions. Experiments on calibration, sensitivity, consistency, and cultural case studies show that MASim reproduces sociocultural phenomena and highlights the importance of multilingual simulation for scalable, controlled computational social science.

As multimodal reasoning improves the overall capabilities of Large Vision Language Models (LVLMs), recent studies have begun to explore safety-oriented reasoning, aiming to enhance safety awareness by analyzing potential safety risks during the reasoning process before generating the final response. Although such approaches improve safety awareness and interpretability, this single-pass think-then-answer paradigm remains vulnerable to contextual or visual jailbreak attacks. This reveals a critical flaw: single-pass reasoning may overlook explicit harmful content in its own output. Our key insight is to exploit this wasted signal through reflection, which can effectively leverage the malicious content revealed in the first-pass reasoning to enable genuine self-correction and prevent unsafe generations. Motivated by this, we propose Think-Reflect-Revise (TRR), a three-stage training framework designed to enhance the safety alignment of LVLMs through policy-guided self-reflection. We first build a Reflective Safety Reasoning (ReSafe) dataset with 5,000 examples that follow a think-reflect-revise process. We then fine-tune the target model using the ReSafe dataset to initialize reflective behavior, and finally reinforce policy-guided reflection through reinforcement learning. Experimental results show that TRR substantially improves the safety performance of LVLMs across both safety-awareness benchmarks and jailbreak attack evaluations, increasing the overall safe response rate from 42.8% to 87.7% on Qwen2.5-VL-7B, while preserving stable performance on general benchmarks such as MMMU and MMStar. The project page is available at https://think-reflect-revise.github.io/.