Safety defenses for large language models (LLMs) are typically trained and evaluated on single-turn prompts, yet real attacks often unfold as indirect, multi-turn probing. To defend against this more nuanced form of deception, we present a unified pipeline that generates realistic multi-turn deceptive question sets via multi-objective genetic prompt optimization with co-evolving mutation operators. We validate this dataset through a human study, which also revealed that early generations yielded the most convincing deception and practical constraints such as adherence filtering and ordering effects. Using this data, we were able to detect deceptive attempts to access prohibited information using simple, explainable geometric signals in embedding space coupled with a lightweight feed-forward classifier. Three geometric features (angular coverage, distance ratio, and linearity) augmented with pairwise similarity statistics led to a compact predictive model that achieved consistently high recall (0.89) across base, reworded, and truncated (three-turn) scenarios, with test-time F1 ranging from 0.74-0.86. The results support a central hypothesis that multi-turn deceptive intent leaves a stable geometric footprint that enables lightweight, transparent screening without expensive end-to-end training. We further discuss responsible uses, limitations, and paths toward larger, more diverse human-evaluated datasets. The primary contribution to artificial intelligence is the multi-objective evolutionary framework for prompt generation, and the engineering application is the deployment of a lightweight geometric detection system for LLM safety infrastructure.

Generative models, including diffusion models, are increasingly used as foundation models and adapted through sequential fine-tuning, making continual learning an essential problem setting. However, continual learning in such generative models remains poorly understood: after a task change, what aspects of the learned distribution are most easily lost, and what replay samples should be prioritized? We address these questions through the modern Hopfield energy. Recent links between modern Hopfield networks (MHNs) and diffusion models allow analyses in MHNs to be transferred to diffusion models. We introduce intrinsic forgetting as an increase in Hopfield energy after the task change. In tractable settings in an MHN, we prove that high-energy, outlier-like samples undergo a larger energy increase than cluster-like samples, implying that samples located in sharp, isolated basins are more forgettable. We further analyze memory replay and show that replay is particularly effective for high-energy samples, enabling an energy-based selection of replay samples. We validate these predictions in experiments on MHNs and two diffusion models under continual-learning settings: Stable Diffusion and a pixel-space DDPM. In these diffusion models, Hopfield energy tracks reconstruction-based forgetting, and replay experiments reveal energy-dependent mitigation of forgetting that is consistent with the MHN analysis.

We study bandit learning in matching markets, where players and arms constitute the two market sides, and the players' utilities are linear in the arm contexts. In each round, new arms arrive with observable contexts. Then, the algorithm matches them to players, aiming to minimize each player's regret against a stable matching benchmark. This contextual structure creates significant complexity: subtle context shifts can slightly alter one player's utility while completely reconfiguring the underlying benchmark, causing large regret spikes for others. We address this in two settings: stochastic contexts, drawn from a latent distribution, and adversarial contexts, which may be arbitrary. For the stochastic case, we introduce a novel minimum preference gap to capture learning difficulty and provide a fully adaptive algorithm with an instance-dependent poly-logarithmic regret upper bound. We also establish matching instance-independent regret upper and lower bounds under a mild distributional assumption. For the adversarial setting, we propose a tractable regret notion that remains valid under arbitrary contexts and achieves an instance-independent sublinear regret bound via an adaptive algorithm.

The world's carmakers are struggling to compete with China Global carmakers are facing a reckoning as US, European and Japanese brands lose ground to Chinese rivals setting the pace not only in electric vehicles, but also in batteries, design and software. The BBC visited factory floors in Beijing and Hefei on the sidelines of Auto China 2026 - the world's largest car show - and found striking levels of automation and software development speed, leaving foreign brands that once dominated the Chinese market struggling to keep up. We have no chance against this, Honda chief executive Toshihiro Mibe told Japanese media after visiting a highly automated factory in Shanghai. Ford chief executive Jim Farley has also warned that Western carmakers, are in a fight for our lives as Chinese rivals expand globally. After decades spent investing in joint ventures with Chinese partners to build vehicles, foreign carmakers are now changing the nature of those partnerships to stay competitive.

One in six young people will not be in education, employment or training within five years unless urgent action is taken, a major review has warned. The education, health and welfare systems are no longer fit for purpose in preparing young people for adult life, said its author former minister Alan Milburn. We are at risk of a lost generation, he warned, with the number of 16 to 24-year-olds out of work, education or training set to rise to 1.25 million by 2031. The first rung of the career ladder has thinned and that for too many young people it is now simply out of reach, Milburn is set to say in a speech later. That places them in a hopeless catch-22 where employers ask for work experience but the opportunities for young people to gain it have narrowed or gone, he will say.

Russian lawmakers have passed a bill to allow trained bank employees to shoot down Ukrainian drones amid an increase in the number of attacks. The draft legislation, which would see banks across Russia install electronic jamming systems while selected employees would shoot down incoming unmanned aircraft, passed in its third and final reading in the lower house Duma on Tuesday, according to the state-run TASS news agency. The bill says the legislation is needed to protect Bank of Russia facilities, including those located in the new constituent entities of the Russian Federation - referring to the four eastern Ukrainian regions that Moscow has announced it has annexed despite not controlling them fully - amid the increasing number of sabotage and terrorist attacks. Under the plan, banks would finance the installation of the equipment on their premises. With banks in almost every town, their incorporation into Russia's air defences could help expand its cover.

Belgium and France have summoned Russia's ambassadors to express anger after Moscow urged foreigners to leave Kyiv in advance of planned "systematic strikes". Brussels and Paris said on Wednesday that Russia's announcement earlier in the week was "unacceptable" and a violation of international law. The spat is unlikely to help smooth the way for the EU to mediate talks to bring the conflict to an end, an arrangement that Russian President Vladimir Putin said on Wednesday he is ready to accept. Germany, the Netherlands, Norway and the European Union summoned Russian envoys on Tuesday following Moscow's warning that foreigners and diplomats should leave the Ukrainian capital before the onset of renewed air strikes . "Threatening embassies is not diplomacy, it is intimidation. And it is a flagrant violation of international law and the Vienna Convention," Belgium Foreign Minister Maxim Prevot said on Wednesday.

The Chinese company is adapting to the demise of Moore's Law, which guides chip production. It could complicate US chip dominance. Tingbo He, president of Huawei's chip-design subsidiary HiSilicon, says her company's engineers have developed a novel way to optimize semiconductors--and she believes it will close the performance gap between Chinese and Western chips over the next few years. Huawei's method, in short, focuses on speeding up computations across chips, circuits, and entire computing systems, rather than squeezing ever-more components onto a single piece of silicon. "We found a new path," He said at the IEEE International Symposium on Circuits and Systems in Shanghai last weekend.

The Israeli military has said it is carrying out air strikes on Hezbollah targets in Tyre in southern Lebanon, after ordering the evacuation of the entire city. The military told residents that it was compelled to act forcefully in Tyre because the Iran-backed armed group was violating a US-brokered ceasefire that began five weeks ago. Earlier on Wednesday, Lebanese media reported a wave of Israeli strikes across the south and the eastern Bekaa Valley, with four people killed in the towns of Choukine and Nabatieh. Hezbollah, which has itself accused Israel of breaching the ceasefire, said it was battling Israeli troops north of the Litani river, about 30km (19 miles) from the border. It came a day after Israel's prime minister announced an expansion of its ground operation following Hezbollah drone attacks on troops occupying part of southern Lebanon and on civilians in northern Israel.

Rescuers in Laos have found five villagers alive inside a flooded cave after they were trapped for a week following heavy rain and landslides. Two people are still missing, rescue teams said. Footage shared by the rescuers showed cave divers crawling through narrow, muddy passageways. The seven people were part of a group of villagers who had gone into the cave in search of gold deposits and wildlife, but could not get out as the cave's entrance was blocked. Could a football match soften North Korea-South Korea relations?