Check your subscription status, update your details and more. Powerful artificial intelligence models built by Chinese companies have gone from inducing widespread panic to being met with a shrug of the shoulders - what changed? When Chinese company DeepSeek released its open-source R1 model in January 2025, it made headlines around the world. The large language model (LLM) was reported to rival some of the most powerful AIs from US companies, but it was completely free for anyone to download. A trillion dollars was wiped off the value of US tech companies and US lawmakers immediately proposed banning it on government devices.

Verity Harding tells WIRED that the US government's nationalistic attitude toward AI is evidence that a worst-case scenario is taking shape. Reports of an artificial intelligence arms race are everywhere-- even in this very publication . But what if that framing is fundamentally dangerous? Between 2016 and 2020, Harding spent her days briefing politicians across the globe, from Barack Obama to Emmanuel Macron, on advances in AI. As the head of global public policy at Google DeepMind, Harding was responsible for mapping out ethical conundrums and potential risks.

Is there any way of fighting back? Anthony "Bingy" Arillotta waited years to become a made man in the Genovese crime family, and when at last the call came in August 2003, he followed directions to the letter. According to sworn testimony, Arillotta was summoned to a steak house in the Bronx, where he was made to hand over his cellphone, beeper, and jewelry before being driven to an apartment building. When he got there, he was taken to a small bathroom and strip-searched for electronic devices. For his big meeting with the boss, he was given a bathrobe to wear. Until recently, only spies and criminals had to worry this obsessively about their private statements being picked up by electronic equipment.

Despite their impressive performance on diverse tasks, neural networks fail catastrophically in the presence of adversarial inputs--imperceptibly but adversarially perturbed versions of natural inputs. We have witnessed an arms race between defenders who attempt to train robust networks and attackers who try to construct adversarial examples. One promise of ending the arms race is developing certified defenses, ones which are provably robust against all attackers in some family. These certified defenses are based on convex relaxations which construct an upper bound on the worst case loss over all attackers in the family. Previous relaxations are loose on networks that are not trained against the respective relaxation. In this paper, we propose a new semidefinite relaxation for certifying robustness that applies to arbitrary ReLU networks. We show that our proposed relaxation is tighter than previous relaxations and produces meaningful robustness guarantees on three different foreign networks whose training objectives are agnostic to our proposed relaxation.

Despite their impressive performance on diverse tasks, neural networks fail catastrophically in the presence of adversarial inputs--imperceptibly but adversarially perturbed versions of natural inputs. We have witnessed an arms race between defenders who attempt to train robust networks and attackers who try to construct adversarial examples. One promise of ending the arms race is developing certified defenses, ones which are provably robust against all attackers in some family. These certified defenses are based on convex relaxations which construct an upper bound on the worst case loss over all attackers in the family. Previous relaxations are loose on networks that are not trained against the respective relaxation. In this paper, we propose a new semidefinite relaxation for certifying robustness that applies to arbitrary ReLU networks. We show that our proposed relaxation is tighter than previous relaxations and produces meaningful robustness guarantees on three different foreign networks whose training objectives are agnostic to our proposed relaxation.

The advancement of general-purpose intelligent agents is intrinsically linked to the environments in which they are trained. While scaling models and datasets has yielded remarkable capabilities, scaling the complexity, diversity, and interactivity of environments remains a crucial bottleneck. Hand-crafted environments are finite and often contain implicit biases, limiting the potential for agents to develop truly generalizable and robust skills. In this work, we propose a paradigm for generating a boundless and adaptive curriculum of challenges by framing the environment generation process as an adversarial game. We introduce a system where a team of cooperative multi-agent defenders learns to survive against a procedurally generative attacker. The attacker agent learns to produce increasingly challenging configurations of enemy units, dynamically creating novel worlds tailored to exploit the defenders' current weaknesses. Concurrently, the defender team learns cooperative strategies to overcome these generated threats. This co-evolutionary dynamic creates a self-scaling environment where complexity arises organically from the adversarial interaction, providing an effectively infinite stream of novel and relevant training data. We demonstrate that with minimal training, this approach leads to the emergence of complex, intelligent behaviors, such as flanking and shielding by the attacker, and focus-fire and spreading by the defenders. Our findings suggest that adversarial co-evolution is a powerful mechanism for automatically scaling environmental complexity, driving agents towards greater robustness and strategic depth.

How is Russia replenishing its military? What is a'coalition of the willing'? How China forgot promises and'debts' to Ukraine How are Europe, the US pulling apart on Ukraine? 'Ukraine is only first': Zelenskyy warns against Putin's expansionist goals Ukrainian President Volodymyr Zelenskyy says the world is in "the most destructive arms race in history" as he calls on the international community to act against Russia now, asserting in his address to the United Nations General Assembly that President Vladimir Putin wants to expand his war in Europe . "Ukraine is only the first, and now Russian drones are already flying across Europe, and Russian operations are already spreading across countries, and Putin wants to continue this war by expanding it," Zelenskyy said on Wednesday at UN headquarters in New York. Moscow has denied the claims of incursions into NATO members' airspace, accusing European powers of levying baseless accusations.

Australia's forthcoming Aukus nuclear-powered submarines have been called the'apex predator of the oceans'. Australia's forthcoming Aukus nuclear-powered submarines have been called the'apex predator of the oceans'. Quantum sensing, satellite tracking and AI are part of an accelerating arms race in detection that should prompt a re-evaluation of Australia's defence strategy Military history is littered with the corpses of apex predators. All once possessed unassailable power - then were undermined, in some cases wiped out, by the march of new technology. " Speed and stealth and firepower," the head of the Australian Submarine Agency, Jonathan Mead, told the Guardian two years ago of Australia's forthcoming fleet of nuclear submarines.

As fears of all-out nuclear war intensify, scientists are sounding the alarm that the decision to launch a catastrophic strike could soon rest not with world leaders, but with a machine. In a stark warning, the Stockholm International Peace Research Institute (SIPRI), an independent group that monitors global security issues, reported that the decades-long decline in global nuclear arsenals has come to an end. Instead, nations are now modernizing, expanding, and deploying their stockpiles at a rapid and alarming pace, signaling the onset of a new, high-tech arms race. While AI and similar technologies can accelerate decision-making during crises, scientists warn they also raise the risk of nuclear conflict through miscommunication, misunderstanding, or technical failure, the report stated. In a nuclear standoff, decision-makers often have only minutes to assess threats and respond.

Artificial intelligence could spark an accidental nuclear war, conflict experts fear. The Stockholm International Peace Research Institute (SIPRI), the world's leading organisation on nuclear assessments, said technologies like AI are aggravating the risk carried with growing global nuclear stockpiles. SIPRI pointed to China's rapidly growing stockpile, from 500 to 600 in a single year, as well as the imminent expiry of the final arms control treaty between the US and Russia, two nuclear-armed nations. The institute's director, Dan Smith, warned: 'One component of the coming arms race will be the attempt to gain and maintain a competitive edge in artificial intelligence (AI), both for offensive and defensive purposes. 'There are benefits to be found but the careless adoption of AI could significantly increase nuclear risk.'