It’s possible that I shall make an ass of myself. But in that case one can always get out of it with a little dialectic. I have, of course, so worded my proposition as to be right either way (K.Marx, Letter to F.Engels on the Indian Mutiny)
Saber-toothed cats' spinal tumors may show how inbred species struggled to survive
Evidence of spinal nerve tumors and developmental abnormalities in saber-toothed cats from the La Brea Tar Pits indicates that the dwindling species was in poor genetic health
Lumbar vertebra HC 12233. The left side of the foramen has been widened, and the blue arrow points to a hole in the bone, thought to have been left behind by a tumor that expanded into this space and eroded the bone. Image by Dr Hugo Schmökel.
The fossilized spines of saber-toothed cats in the La Brea Tar Pits reveal evidence for an unexpectedly high prevalence of rare spinal nerve tumors, suggesting that the species became inbred just before extinction. These tumors, which cause significant pain and reduced limb function, could have made the cats particularly vulnerable to the lure of prey animals stuck in the tar pits.
“A trapped prey animal is very attractive for a predator in pain,” said Dr Hugo Schmökel of the Evidensia Academy Sweden and the University of Zürich, lead author of the article in Frontiers in Veterinary Science. “A wounded or sick predator must rely on carcasses for surviving, so they’ll take bigger risks to get to them.”
A death trap
At the La Brea Tar Pits in California, crude oil seeps up through cracks in the ground in the form of ponds of sticky black tar which can trap animals that walk into them. The distress calls of trapped prey attract predators, which then become trapped themselves. During the Pleistocene, this included saber-toothed cats — Smilodon fatalis.
“Smilodon was an ambush hunter, hiding in the lush vegetation of the late Californian Pleistocene, jumping on prey and wrestling it down, until a fatal stab ended the hunt,” said Schmökel. “But scavenging certainly also happened.”
Well-preserved specimens of Smilodon held by the La Brea Tar Pits Museum allow scientists to look for evidence that could help us understand the species’ extinction. Schmökel and his colleagues investigated Smilodon’s spines. They identified 3,722 vertebrae, representing at least 849 adults, in the museum’s collection, and examined these for pathological changes.
They found numerous cases of developmental abnormalities: 32 vertebrae with incompletely developed arches, 48 fused ‘block’ vertebrae, and 226 ‘transitional’ vertebrae, where vertebrae form incorrectly — for example, cervical vertebrae that form like thoracic vertebrae, with a rib.
Even more strikingly, the scientists found widened foramina, holes which allow nerves to travel through the spine, in three vertebrae. Widened foramina with a smooth remodeled surface are a very strong indication that the individual was affected by a spinal nerve tumor: the tumor grows and exerts pressure on the bone over time, increasing the size of the hole.
“Dogs which are brought to my clinic with a spinal nerve tumor mainly have signs of pain, and many human patients with a spinal nerve tumor complain of pain,” said Schmökel. “This was probably also the case in Smilodon. In advanced cases without treatment the nerve and the affected limb eventually lose function, making hunting very difficult.”
Spinal nerve tumors are extremely rare in both humans and modern-day animals. Human cases are caused by specific genetic mutations, either spontaneous or inherited. Doctors see approximately 0.22-0.38 cases for every 100,000 humans. By contrast, the presence of three cases — the affected vertebrae came from three different pits, so must have belonged to different animals — means that the estimated prevalence rate at La Brea Tar Pits is a staggering 353 cases for every 100,000 Smilodons.
Small population, big risks
We can’t be sure why Smilodon was so severely affected by spinal nerve tumors. But the high prevalence of developmental abnormalities leads the team to suspect that the species had become inbred, although this can’t be proven without genetic analysis. As their numbers began to fall, so that fewer potential mates were available, harmful genetic variants could have become concentrated in the population, increasing the likelihood that each litter of kittens inherited health problems. The same spinal malformations have been documented in modern-day inbred grey wolf populations.
The scientists point out that the tumors themselves could have increased the chances that the cats ended up in the tar pits: pain and difficulty hunting might have led them to take greater risks than healthy animals. But for Schmökel, the evidence of inbreeding is a wake-up call regarding our treatment of modern megafauna.
“Reduced fitness and smaller litters don’t help a population under stress,” said Schmökel. “But inbreeding is not the cause of the crash of populations, it is a medical consequence. Loss of habitat, pollution, hunting or poaching, and in some areas today climate change, are the cause of the fall in numbers of wild cats. We can assume that the same happened to Smilodon at the end of the Pleistocene.
“We must take care of living wild animals. We humans reduce their living space, hunt them, pollute them. And when there are only a few left, we start, maybe, to act. But by then these animals are probably affected by inbreeding and saving them is more difficult.”
Lumbar vertebrae HC 12288. The white arrow on the CT scan shows the pathologic widening of the left foramen of the fourth and fifth lumbar vertebrae, and the yellow arrow shows where the third and fourth lumbar vertebrae have fused together. Image by Dr Hugo Schmökel.
Wildlife protection frameworks in both the EU and the UK need stronger and more consistent implementation – and must recognise animals as “individuals capable of experiencing suffering”, rather than mere ecological assets.
This is the argument from authors of a new peer-reviewed study, which, in providing the first comprehensive comparative examination of EU and UK wildlife legislation in the post-Brexit era, exposes a disconnect between ambitious policy goals and practical implementation.
The research team, environmental lawyers Dr Caroline Cox and Dr Meganne Natali of the University of Portsmouth, reveal significant shortcomings in wildlife protection frameworks across Europe and the United Kingdom, despite decades of legislative development and billions in conservation investment.
“Our study finds that while both the EU and the UK have developed complex legal structures for wildlife protection, neither system delivers a coherent or fully effective framework,” Drs Cox and Natali explain.
“In the EU, wildlife protection remains fragmented, selective, and exception-based, with species safeguarded only when expressly listed and protection frequently weakened through exception that allows a national, local or regional administration in an EU member state to deviate from a given regulation (derogations) and political compromise.
“In the UK, outdated legislation and weak enforcement further undermine conservation outcomes.”
Key findings
European Union framework under scrutiny
The research identifies fundamental contradictions within the EU's wildlife protection system, despite its reputation as one of the world's most comprehensive:
Only 16% of habitats listed under the Habitats Directive (an EU environmental legislation designed to protect endangered species and habitats) are currently in “favourable condition”
53% of bird species assessed between 2013-2018 showed “unfavourable conservation status” (when a species or habitat is not considered to be in a healthy, secure, and sustainable condition for the long term).
The EU's protection system operates through "exceptions rather than universality," (the way in which EU wildlife laws protects only selected species and habitats, rather than providing protection to wildlife generally) leaving countless species without legal recognition.
UK post-Brexit challenges
The Wildlife and Countryside Act 1981, Britain's cornerstone wildlife legislation, faces mounting criticism:
Nearly one in six of the UK's 10,000+ surveyed species risk extinction
Only 14% of important wildlife habitats are in good condition
Wildlife crime conviction rates remain significantly below average for all crimes
The Act's five-yearly review system leaves species vulnerable to political shifts and ministerial priorities
An anthropocentric logic
A key problem baked into both of the frameworks is a certain anthropocentric logic, the paper claims. Under this, wildlife is protected because of its value to humans—for example, via ecological services, agricultural balance or landscape aesthetics—rather than for the wildlife themselves.
“Even where animal sentience is recognised at treaty level,” the authors note, such as in EU primary law and the UK’s Animal Welfare (Sentience) Act 2022, “this recognition has not been operationalised within biodiversity law.”
According to the lawyers, the benefit of environmental protections regarding wild animals specifically as sentient beings is that it introduces ‘ethical continuity’ into the legal framework; at present, protection is all too conditional and reversible.
“Species are protected when they serve ecosystem functions of policy objectives—and downgraded when they become politically inconvenient," the authors add.
The European Wolf controversy highlights a “fragile” protection system
This impermanence is exemplified by the EU’s 2024 decision to downgrade the protection status of wolves from ‘strictly protected’ to ‘protected’, granting greater flexibility in the management of wolf populations, including via culling.
This move has been framed as a response to increasing wolf populations across Europe, and a corresponding increase in conflicts with farmers and hunters.
Although the wolf has been a significant conservation achievement for the EU—with numbers having increased by nearly 60% in a decade—studies have cautioned that wolves have not yet achieved the benchmark of a genuinely favourable conservation status.
Furthermore, Cox and Natali note, the European Commission’s prior analysis in 2023 did not support a reduction in the protection level; it also acknowledged that coexistence measures are more effective at protecting livestock from wolves than culling.
The decision to downgrade the protection of wolves, the authors note, was also “marred by procedural shortcomings”, including a restricted public consultation and a lack of transparency around data on livestock losses and wolf behavior; accompanied by pressure from agricultural and hunting lobbying groups; and surrounded by controversy as to whether the predation of European Commission President Ursula von der Leyen’s pony by wolves influenced the policy shift.
“The wolf downgrade demonstrates how fragile protection can become under pressure,” the researchers note.
Another issue in the EU is that wildlife law only protects those species that are explicitly listed in the annexes of the Habitats and Birds Directive – based on scientific assessments of rarity, conservation status and ecological value at the time of drafting.
“Species outside the annexes receive little to no protection,” Dr Natali says.
“Member States can technically comply while limiting the practical scope of conservation. The result is a framework that appears harmonised, but in reality remains fragile and uneven in application.”
Tightening regimes for greater protection
As for how they would like to see protection frameworks improved in both the EU and the UK, Cox and Natali highlight three areas for improvement.
“First, derogation regimes must be tightened,” they say. “Protection cannot remain structurally dependent on broad ‘overriding public interest’ clauses.”
Secondly, the enforcement of protective measures must be strengthened. As the researchers note: “Legal ambition without monitoring and prosecutorial follow-through produces symbolic protection.”
The third would include enhanced cooperation and coexistence-based approaches. “We advocate stronger cross-border cooperation, better integration of wildlife conservation across policy sectors, and the promotion of human–wildlife coexistence strategies rather than conflict-based management,” Dr Natali adds.
Implications for global conservation
The study's findings extend beyond Europe, offering lessons for wildlife governance worldwide. As biodiversity loss accelerates globally, the research underscores the urgent need for legal frameworks that balance human interests with ecological integrity and ethical responsibility.
The research team argue that wildlife law must place animal sentience at the heart of conservation frameworks.
“Without that integration, biodiversity law remains ethically incomplete and politically unstable.
“The true measure of environmental law lies not only in its capacity to preserve species, but in its willingness to govern our shared landscapes with justice, empathy, and foresight,” the authors conclude.
With their initial study complete, the researchers are now moving to develop practical recommendations for implementing improvements to existing wildlife protections – examining how these laws could integrate sentience recognition without collapsing into pure welfare regulation – aiming for a coexistence-based framework that bridges biodiversity governance and animal law.
Summary of the main findings from 75 global biodiversity experiments. Biodiversity effects on productivity were strongest under extreme drought in more-arid grasslands, driven by increased complementarity, whereas comparable context dependence was not detected in forests or under heat extremes.
Biodiversity does not enhance ecosystem productivity equally across all ecosystems under climate extremes. During years of extreme drought, drier grasslands showed the strongest positive effects of biodiversity on productivity.
When extreme drought strikes, drier grasslands receive the greatest productivity benefit from biodiversity. By contrast, forests did not show the same context-dependent pattern under drought, according to a new global synthesis of 75 biodiversity experiments. Researchers from YOKOHAMA National University published their results in Nature Ecology & Evolutionon July 15.
Biodiversity has been studied long before it got its catchy name, though fully exploring all it entails is a never-ending process. Through other studies, it has already been established that biodiversity plays a significant role in an ecosystem’s productivity. What is lesser known is where this biodiversity matters most when climate extremes are taking their toll.
“Our ultimate goal is to move away from a broad ecological insight to a practical basis for climate adaptation,” said Takehiro Sasaki, professor of the Faculty of Environment and Information Sciences at YOKOHAMA National University and first author of the study.
With intense heat waves and droughts becoming more commonplace, this question becomes a crucial one in the space of climate science. A global synthesis of 75 biodiversity experiments sought to answer this question, along with whether these benefits persist, weaken or intensify under times of drought or extreme heat, and if the benefits of biodiversity apply equally to different ecosystems.
In more-arid grasslands, plant diversity had its strongest positive effect on productivity during years of extreme drought. This effect was driven mainly by stronger complementarity among species, consistent with species contributing in more functionally distinct and/or mutually supportive ways under water limitation. In less-arid grasslands, by contrast, drought was associated with stronger selection effects, indicating a greater contribution from a few highly productive species.
Forests did not show comparable context dependence under extreme drought, although this does not mean that biodiversity is unimportant in forests. Heat extremes likewise did not produce clear context-dependent changes in biodiversity effects across ecosystem types or aridity gradients. Across both grasslands and forests, soil nutrient conditions did not detectably modify biodiversity effects under either drought or heat extremes, suggesting that water limitation may become a more important constraint on productivity than soil nutrient supply as climatic stress intensifies.
The study synthesized data from 75 biodiversity experiments in grasslands and forests spanning broad climatic gradients, with experiment durations ranging from 2 to 23 years. These data were linked to long-term daily precipitation and maximum-temperature records, as well as site-level aridity and soil data. The analysis assessed whether aridity and soil nutrient conditions modified biodiversity effects under drought and heat extremes
Results show that biodiversity effects on productivity were strongest under extreme drought in drier grasslands, whereas forests show no comparable context dependence under the same conditions.
Further exploration to fully unpack the question of where biodiversity is going to make the biggest difference when climate extremes hit is necessary. Researchers hope to improve their forest evidence by increasing the duration of studies, improving more sensitive indicators of drought for forest ecosystems, in addition to testing whether the effects of biodiversity are slower to emerge in forests than in grasslands. Looking into the tree health indicators and the crown condition of forests years after extreme climatic anomalies might be better indicators of the ecosystem’s health and the weight the biodiversity of the forest carries.
Another goal of this research is to make the science more predictive for use in geographical blind spots in biodiversity experiments. This type of insight would help prepare ecosystems to better adapt to a warming climate, and ideally make conservation efforts a top priority when it comes to preserving the varied environments present on this planet.
Takehiro Sasaki, affiliated with both the Graduate School of Environment and Information Science and the Institute for Multidisciplinary Sciences at YOKOHAMA National University, together with Yuki Iwachido of the Graduate School of Environment and Information Science and Nico Eisenhauer of the Institute for Multidisciplinary Sciences at YOKOHAMA National University, contributed to this research.
The Ministry of Education, Culture, Sports, Science and Technology of Japan, Tottori University, Deutsche Forschungsgemeinschaft, NSF Biodiversity on a Changing Planet Program and NSF Long-Term Ecological Research Prgroam, Consejo Nacional de Ciencia y Tecnologia, Swiss National Science Foundation, Margarete-von-Wrangell Fellowship of the Ministry of Science, Research and Arts Baden-Wurttemberg and the European Social Fund, NSF Awards, NSERC DG grant, EXCELLENTIA project, the German Research Foundation, Swedish Research Council Formas and the Agence Nationale de la Recherche made this research possible.
###
YOKOHAMA National University (YNU) is a leading research university dedicated to academic excellence and global collaboration. Its faculties and research institutes lead efforts in pioneering new academic fields, advancing research in artificial intelligence, robotics, quantum information, semiconductor innovation, energy, biotechnology, ecosystems, and smart city development. Through interdisciplinary research and international partnerships, YNU drives innovation and contributes to global societal advancement.
Developed by the Luxembourg Institute of Science and Technology (LIST) and the University of Luxembourg’s Interdisciplinary Centre for Security, Reliability and Trust (SnT), the tool lets any organisation build a customised environment for testing if an artificial intelligence system is trustworthy and compliant. This tool lays the groundwork for a shared European infrastructure for AI assessment.
Artificial intelligence adoption is accelerating across Europe’s economy and society, and with it the urgency of ensuring that AI systems can be trusted. Organisations cannot unlock the full value of AI unless people trust it: if users doubt every output, or avoid AI altogether, the expected gains never materialise. The EU AI Act and other regulations also impose clear obligations on AI providers to demonstrate trustworthiness. Meeting those obligations today is slow, costly and largely manual. The AI Assessment Sandbox Configurator aims to remove such bottlenecks and enable rigorous AI testing at scale for the growing number of AI agents that are being deployed by organisations.
Customised AI testing environments for public and private organisations
Ensuring an AI tool is trustworthy and compliant with regulatory requirements involves converting policy requirements into controls and metrics, identifying the relevant tests available, integrating them, reporting test results and preparing for audits. All of this remains a painstaking, largely manual process. The AI Assessment Sandbox Configurator is open source, allowing any public or private organisation to accelerate the creation of customised AI testing environments that can be deployed anywhere — on premises, on private or sovereign clouds — making it suitable for organisations with strict data residency or security requirements, for example financial firms.
With the AI Assessment Sandbox Configurator, users gain access to a curated and extensible catalogue of AI tests, controls and datasets, from which they can select those relevant to the specific requirements of their use case. The tool then generates a customised testing environment including a collaborative dashboard to visualise and assess the results, a report generator and temper-proof evidence for auditing.
It is aimed at any organisation — large enterprises, public institutions, start-ups and SMEs — that wants to deploy AI responsibly in cooperation with Competent Authorities and notified bodies.
Built on years of hands-on AI testing experience
The Configurator is a natural next step in SnT’s and LIST’s work on AI assessment.
For the University of Luxembourg’s SnT, the Configurator builds on almost a decade of extensive research partnerships across industry and the public sector. These partnerships have resulted in research papers published in scientific journals and conferences for AI that are in the top ten percent globally. Since a partnership with BGL BNP Paribas that began in 2017, SnT has also worked with Spuerkeess, the Luxembourg Ministry of Digitalisation, and more recently the European Stability Mechanism (ESM). The research team is also currently collaborating with the National Bank of Greece on a Horizon Europe project, to assess selected AI systems. This industry perspective informed the creation of the AI Sandbox tool.
Maxime Cordy, Assistant professor in Software Engineering for AI Systems and SnT’s project lead, said:
“In our work with partners we experienced first-hand the challenge to integrate cutting-edge technology into business-critical operations. This enabled us to produce breakthrough scientific research and in doing so we realised we could use our expertise to contribute an open-source tool to ease AI adoption. AI adoption requires both speed and control, and the AI Assessment Sandbox Configurator enables organisations to integrate AI in their processes in a trustworthy manner, supporting them in their global competitiveness.”
For several years, LIST has operated its own AI Sandbox, a hands-on testing environment where organisations evaluate AI models for robustness, fairness, bias and regulatory compliance. That Sandbox has been put to work with organisations including Banque Internationale à Luxembourg, the City of Luxembourg, and Mistral AI.
Francesco Ferrero, Leader of the Flagship Initiative on Artificial Intelligence and Head of the Human-Centred AI, Data and Software Research Unit at LIST, said:
“With the LIST AI Sandbox, we showed that rigorous, independent AI testing is achievable in practice. The Configurator is the logical next step: it packages that expertise into an open tool so that any organisation, anywhere, can stand up its own assessment environment without starting from scratch.”
Made in Luxembourg, built for Europe’s sovereignty
Luxembourg AI Factory sees the Configurator as the foundation of something bigger: a shared European resource, co-developed by regulators, researchers, certification bodies, and companies across Member States. The tool also positions Luxembourg as a hub for international companies, particularly in sectors such as finance that wish to deploy AI systems in Europe and need a trusted, sovereign environment in which to assess them.
Over time, the library of tests and controls is expected to expand into specialised offerings for industries such as healthcare, finance, manufacturing and others. Several pilots are already underway across Europe, and the underlying research has been published in several peer-reviewed scientific papers.
Stefano Pozzi Mucelli, Head of Ecosystem Innovation Projects at Luxinnovation, concluded:
“AI adoption only delivers value when AI systems can be trusted. The AI Assessment Sandbox Configurator is a natural extension of Luxembourg AI Factory’s service catalogue: while our services already guide companies from assessment through to deployment, this tool enables any organisation to map regulatory requirements to concrete assessment metrics and generate audit-ready reports on its own infrastructure. It allows us to better serve companies of all sizes, from SMEs to large enterprises, by making trustworthiness practical rather than theoretical. This is how Luxembourg is positioning itself at the forefront of AI in Europe: by treating trust not as a barrier, but as the enabler that helps AI innovation reach the market and its users.”
The relationship between countries' climate vulnerability and explicit SDG references within their climate commitments (left). The same relationship following the AI-driven identification of implicit SDG references, revealing a significantly deeper integration between climate action and sustainable development (right).
An international team including researchers from the University of Alicante (UA) and the Universitat Politècnica de València (UPV) has used artificial intelligence to analyse the climate commitments submitted to the United Nations by 158 countries. Their conclusion is stark: profound inequalities persist within global climate planning.
The paper, published in the journal Nature Communications, concludes that high-income nations focus their climate commitments on health, technological transitions and emissions reduction. Conversely, low- and middle-income countries tie climate action to immediate survival challenges – such as access to water, energy, food security and natural resource management.
Researchers analysed Nationally Determined Contributions (NDCs) – the periodic climate action plans submitted under the Paris Agreement – using advanced generative AI models. This technology identified implicit connections between national climate measures and the UN Sustainable Development Goals (SDGs).
Analysing all this information allows us to understand countries' priorities, the risks they consider most important, and where potential inconsistencies or blind spots exist before new decisions are adopted, as explained by Javier García Martínez, one of the authors and a professor at the University of Alicante. The findings are particularly time-sensitive as governments worldwide prepare their next round of climate pledges for 2035.
The study found that over half of the countries analysed do not explicitly mention the SDGs in their pledges. Furthermore, pillars of sustainable transition such as education and gender equality are poorly represented across the board, regardless of a nation's income level.
According to Sergio Hoyas, a professor at the Universitat Politècnica de València (UPV) who participated in the study, "These results highlight critical misalignments between the climate agenda and the sustainable development goals driven by the United Nations".
AI as an ally against cimate change
Beyond diagnosing national priorities, the authors argue that generative AI can serve as a powerful tool to evaluate the quality and coherence of climate policies before they come into effect. The research team suggests this analysis will help governments, international agencies and funding bodies identify genuine priorities, improve resource allocation and prevent future climate strategies from worsening existing global inequities.
"At a time when the international community is debating how to accelerate climate action and finance the energy transition, this study offers an unprecedented map of the concerns, aspirations and contradictions within national climate plans worldwide," concluded UPV professor Alberto Conejero.
The interdisciplinary project brought together Teaching and Research Staff (PDI) from the University of Alicante's Department of Inorganic Chemistry and the UPV's Institute of Pure and Applied Mathematics, alongside experts from the KTH Royal Institute of Technology, the University of Oxford and the University of Michigan.
The TechBrief, “Agentic AI: Autonomy, Opportunities, and Challenges of Action-Taking AI Systems,” examines AI systems that plan and execute multi-step tasks toward a user-defined goal. Such agentic systems are being rapidly adopted by enterprises and consumers. A 2025 survey of more than 500 technology leaders found that 48% are already deploying or adopting agentic AI. The TechBrief finds that this acceleration is outpacing the legal, regulatory, and technical frameworks designed to govern it.
AI systems are increasingly browsing the web, executing code, managing files, and sending messages without step-by-step human approval, raising new risks in the process, according to a new TechBrief from the Association for Computing Machinery’s Technology Policy Council (TPC) on the rise of agentic AI.
The TechBrief, “Agentic AI: Autonomy, Opportunities, and Challenges of Action-Taking AI Systems,” examines AI systems that plan and execute multi-step tasks toward a user-defined goal. Such agentic systems are being rapidly adopted by enterprises and consumers. A 2025 survey of more than 500 technology leaders found that 48% are already deploying or adopting agentic AI. The TechBrief finds that this acceleration is outpacing the legal, regulatory, and technical frameworks designed to govern it.
The brief arrives as governments are beginning to respond. On May 1, 2026, CISA and five allied national cybersecurity agencies published the first coordinated multinational security guidance specifically targeting agentic AI, just 42 days after the White House released its National Policy Framework for Artificial Intelligence on March 20. Yet neither effort fully resolves the accountability questions agentic systems raise.
“Many people are rapidly adopting agentic AI systems for their businesses and personal lives. They know that these systems can cause great harm when they misbehave, but the short-term advantages of deploying them and hoping for the best are nearly impossible to resist,” said Simson Garfinkel, Chief Scientist at BasisTech and Chair of the ACM TechBriefs Committee. “These systems can offer tremendous advantages to their users, but anyone deploying them today is taking on real risk with very little legal protection. When something goes wrong with a system that takes actions on a user’s behalf, it can be genuinely difficult to determine who is responsible. Existing law simply doesn’t answer this question.”
The TechBrief identifies four key policy dimensions where existing frameworks fall short:
Legal liability without a clear accountable party: When an AI agent causes harm, as in a documented case where one deleted a company’s entire production database, responsibility may fall to the model provider, the framework developer, the person or company that deployed the agent, or the end user. No person made the decision, yet harm was done. No case law currently exists that can help resolve liability.
Serious and underappreciated security risks: Because Large Language Models (LLMs) process text as both data and commands, agentic systems cannot reliably distinguish legitimate content from embedded malicious instructions. Documented incidents include an AI agent that exposed private Slack data after processing a message containing hidden instructions, and consumer agent marketplaces found to contain malicious extensions reaching hundreds of thousands of users.
Lack of consumer transparency and recourse: Users often cannot determine what systems an agent can access, what actions it can take without confirmation, or how to revoke its permissions. No standardized data format currently exists to disclose or control agent authority, a concern the brief flags as particularly acute in high-stakes settings such as healthcare.
Workforce disruption outpacing evidence: A 2025 Gartner survey found that 55% of supply chain leaders expect agentic AI to reduce entry-level hiring. Yet the productivity claims driving those decisions have not been independently verified at scale, and the long-term effects on skill formation and labor markets remain unmeasured.
“Who is legally responsible when autonomous systems cause harm?” Garfinkel continued. “Today’s license agreements just point fingers elsewhere. But if there is no underlying technology that can make these systems reliably follow the policies they are given, that finger-pointing may not be legally binding. This is not a problem specific to the US, Europe, or any single Asian country. We need to work on both law and technology to provide strong assurances to businesses and consumers throughout the entire industrialized world.”
The TechBrief concludes that addressing these challenges will require defined authentication and delegation standards, robust audit trails, standardized consumer disclosures, and sector-specific guidance in areas such as healthcare, financial services, and critical infrastructure, where existing law assumes a human decision-maker.
ACM’s TechBriefs are designed to complement ACM’s activities in the policy arena and to inform policymakers, the public, and others about the nature and implications of information technologies. Earlier ACM TechBriefs have covered topics such as vibe coding, buying vs building LLMs, automated speech recognition, governmental digital transformation, accessibility, and generative artificial intelligence among others.
About the ACM Technology Policy Council ACM’s global Technology Policy Council sets the agenda for global initiatives to address evolving technology policy issues and coordinates the activities of ACM’s regional technology policy committees in the US and Europe. It serves as the central convening point for ACM’s interactions with government organizations, the computing community, and the public in all matters of public policy related to computing and information technology. The Council’s members are drawn from ACM’s global membership.
About ACM ACM, the Association for Computing Machinery, is the world’s largest educational and scientific computing society, uniting computing educators, researchers, and professionals to inspire dialogue, share resources, and address the field’s challenges. ACM strengthens the computing profession’s collective voice through strong leadership, promotion of the highest standards, and recognition of technical excellence. ACM supports the professional growth of its members by providing opportunities for life-long learning, career development, and professional networking.
UW researchers built AI agents that quickly estimate electronic devices’ carbon footprints
If you shop on Google Flights, you get a quick comparison for different itineraries: One flight’s carbon emissions may be average, while another’s are 14% higher. But if you go shopping for a new laptop, you likely won’t find quick, comprehensible information on different models’ sustainability bonafides, despite thenotable environmental impactsof producing and discarding electronics. In part, that’s because understanding a device’s emissions is difficult and time-consuming, even for experts.
University of Washington researchers developed an artificial intelligence system that automatically estimates the environmental impacts of making different electronic devices. The system uses AI agents — programs that perform tasks autonomously — to comb through publicly available data and conduct life cycle assessments, or LCAs. The system achieves an average error rate of 5%-19%, similar to the accuracy of LCAs conducted by experts.
“Recent studies have shown that people are willing to pay more for more sustainable devices,” said senior authorVikram Iyer, a UW assistant professor in the Paul G. Allen School of Computer Science & Engineering. “So there’s growing demand for this information. But a phone, for example, is made of hundreds of chips and other components, and producing each of those causes varying amounts of emissions. Since that data isn’t public or sometimes not even measured, human experts can spend days, even months manually gathering information for LCA. Instead we designed multiple AI agents that work together to automatically find this data and produce comparable estimates in about a minute.”
AI agents have recently grown increasingly capable of performing complex tasks. Today's agents can search the web and pull information about electronic parts from product descriptions, images and documents.
“Some ofour previous researchmade me curious about how LCA experts perform environmental assessments — and whether that process could be automated,” said lead authorZhihan Zhang, a UW doctoral student in the Allen School. “Sowe interviewed LCA expertsto understand the bottlenecks firsthand, and then built a system that emulates these interactions with two AI agents. Each of them mimics different roles in the LCA process.”
One agent acts as a sort of analyst, defining what information needs to be gathered and how it will fit together. It also reviews results for accuracy. The second agent is more like an engineer. It scrapes publicly available data for information on an electronic device’s components. That might entail sifting through spreadsheets, or looking up images of the insides of devices and taking chip information from them — including from sources not typically used for LCAs, such asFCC databasesand posts oniFixit.
The two agents work in a loop. The first sets the scope, the second gathers information. The first then looks that information over and might send the second agent searching again, and so on. The agents then referenceLCA databasesto convert the complete list of parts to carbon estimates.
The team also developed a new method to bypass this detailed data collection and directly estimate carbon footprints. For common devices like laptops and smartphones with publicly available carbon footprint reports, they found that products with similar specs like screen size and processors clustered around similar carbon values, because only a handful of companies make specialized parts for all these devices. So an unknown device's footprint can be represented as a weighted average of similar products.
They also use this to estimate the carbon for materials not in LCA databases. For example, a new type of sustainable plastic could be estimated based on plastics with similar properties and chemistry.
“We tried this ‘nearest-neighbors’ approach and found that for materials, it’s actually better than the standard approach of a human picking the single closest entry,” said Zhang. “When estimating missing emissions factors in a test, the average error for our method was 23%. Human experts had an average error of 143%.”
The authors note that while the aim of the system is to help reduce carbon emissions overall, running AI models requires energy, so they’ve taken several steps to mitigate its impact. They use small AI models that aren’t as energy-intensive as general-purpose models. They also start the process by running a search to see if the device’s estimated emissions have already been calculated. If so, it can stop there. If the system does need to call its AI models repeatedly, estimating a device’s carbon footprint is currently on par with the emissions generated by brewing a cup of tea.
The team plans to collaborate with companies in the future to help automate their workflows.
“A lot of big companies have sustainability teams that perform these LCAs,” Iyer said. “Our hope is that automating this will actually free up their time, so they can spend their time reducing the carbon footprint of the products themselves, instead of hunting down elusive stats.”
Co-authors includeAlexander Metzger, a UW student in the Allen School;,Felix Hähnlein, a UW postdoctoral researcher in the Allen School;Zachary Englhardt, a UW doctoral student in the Allen School;Shwetak Patel, a UW professor in the Allen School;Yuxuan Meiof Wesleyan University, who completed this research as a UW doctoral student in the Allen School;Tingyu Chengof the University of Notre Dame;Gregory D. Abowdof Northeastern University; andAdriana Schulzof Brown University, who completed this research as a UW assistant professor in the Allen School.
This research was funded by Amazon Research Awards and the National Science Foundation. Zhang was supported by theGoogle PhD Fellowship.
Sheets of molybdenum ditelluride crystals, when stacked on top of one another in a specific way, create the complex lattice structure seen above. In a new study, materials scientists at the University of Washington used artificial intelligence to simulate huge stacks of these sheets, producing new quantum phenomena that were not present at smaller scales.
Quantum materials are a class of exotic materials with special properties that are governed by quantum mechanics rather than classical physics. Those properties — like superconductivity, entanglement and unusual forms of magnetism — often originate in the tiny repeating patterns of atoms inside crystals, but through clever engineering they can be observed and controlled at a more human scale. Quantum materials are helping to power the quickly growing field of quantum computing, and could find their way into future generations of energy-efficient electronics.
Designing new materials from the atomic scale up, however, requires intense modeling and simulation. Some materials may appear ordinary when viewed as small clusters of atoms, yet reveal new and useful properties when their atomic building blocks repeat and interact over larger distances. Researchers must be able to accurately predict behaviors at large scales in order to find materials with practical applications — otherwise designing new materials is a slow and costly trial-and-error process.
In the past 50 years, supercomputers have helped materials scientists solve some of those thorny prediction problems, but two recent studies from the University of Washington demonstrate how newer computing techniques can help researchers sniff out promising quantum materials to pursue. The first study, published June 2 in the Proceedings of the National Academy of Sciences, shows how researchers can use artificial intelligence to simulate dozens of sheets of atoms stacked in intricate patterns, a process that produces complex and potentially useful quantum behaviors. The second study, published June 8 in Nature Communications, shows how quantum computers can create a self-improving design loop by discovering new materials that could themselves be components of future quantum computers.
“What is exciting is that AI and quantum computing are beginning to change not just what problems we can solve, but how we do research,” said Ting Cao, a UW associate professor of materials science and engineering and the senior author of both studies.
These two new tools — AI and quantum computing — are complementary in that they each excel at a different kind of simulation problem. With the right training, an AI model can act as a fast and relatively inexpensive surrogate of a supercomputer, extrapolating the behavior of huge material systems from a relatively small dataset. Cao and collaborators used this approach to stack virtual sheets of atoms on top of one another over and over — a process that created completely new phenomena that were absent on a smaller scale, but would have been impractical to model by traditional supercomputing. From there, researchers can try to make the most promising materials in the lab to prove out the simulations.
Quantum computers, on the other hand, are essentially powered by the same quantum phenomena — like entanglement — that Cao and other materials researchers want to study. Such phenomena can be difficult to simulate using traditional computers or AI systems, but quantum computers are naturally suited to the task. In the study, Cao and his team used a quantum computer to study an exotic phase of matter known as a Laughlin state.
Moving forward, Cao and his team plan to further build out their datasets and eventually develop models that can simulate a much wider range of materials. They also hope to combine their AI and quantum computing systems into a more powerful and flexible hybrid tool.
“The next step is to bring these tools together,” Cao said “We can use AI to guide quantum simulations, and quantum computers to generate new data and insights that improve AI models.”
“We are at the start of a new era,” said Di Xiao, UW professor and chair of materials science and engineering and co-author of both studies. “Our field is fundamentally changing. Things that were literally impossible a couple of years ago are now becoming routine. And we are only beginning to see what AI and quantum computing will make possible for quantum materials.”
The first study was led by Yueyao Fan, a UW doctoral student of materials science and engineering. The second study was led by Lingnan Shen, a UW doctoral student of physics. A complete list of authors is included with the studies.
The authors acknowledge the support of Amazon and the Department of Energy.
Detecting phishing campaigns when malicious content is cloaked. PhishLumos is triggered when it finds that content is cloaked and uses infrastructure clues to uncover details of the whole phishing campaign.
Tokyo, Japan – Researchers from Tokyo Metropolitan University have created a new paradigm for identifying online phishing campaigns. Their new system, PhishLumos, is triggered when links show signs of concealing information, and looks for clues in the “infrastructure” of the website to uncover the whole campaign of which the site is only a tiny part. Real-world testing showed detection which was 8 days faster than an expert, with 190,000 URLs detected over 6 months.
Phishing is a rampant form of cybercrime. Criminals impersonate trusted entities like banks or employers to get victims to share sensitive information, click malicious links, or install harmful software. Less digital savvy people are at particular risk, which not only widens the digital divide but erodes trust in essential digital institutions.
This is why researchers have been looking for ways to shut down phishing campaigns. However, they face severe challenges. For example, most existing approaches involve analyzing individual, suspicious links on the web, or Uniform Resource Locators (URLs). While machine learning and deep learning approaches have helped realize increasingly sophisticated programs that can assess content for veracity, cybercriminals can generally produce far more malicious links in the same time it takes to identify and shut down one site. Malicious content generation is also becoming cleverer; cloaking technologies can help fool scanners, leading to more malicious content making it in front of potential victims.
Now, researchers are looking for a paradigm shift. In recent work, a team led by Associate Professor Daiki Chiba from Tokyo Metropolitan University has adopted a new approach. Rather than trying to label single links as good or bad, they look for signs of cloaking as a starting point for a whole, automated investigation to identify the whole phishing campaign associated with a malicious actor. Their system, PhishLumos, is not evaded by withheld content, but triggered instead. Once activated, it will look for clues in the “infrastructure” of the URL, like which Internet Protocol (IP) numbers are involved, and which network connections are used. These help map out the whole campaign of URLs involved in the same phishing project, not simply as a big list of URLs, but a so-called Knowledge Base (KB) graph which describes how the campaign works.
Looking at 103 real phishing campaigns, PhishLumos was able to achieve detection 8 days faster than experts on average. In real-world tests, given 600 seed URLs as starting points, the rules that it uncovered led to the discovery of over 190,000 new links of which 92% were later flagged as malicious. Importantly, it significantly outperformed so-called “content-centric” approaches which go through website content instead of infrastructure clues.
Online services are already an indispensable part of modern society, so bad actors can cause widespread, irreparable harm to society. Projects like PhishLumos are an essential part of making sure that the benefits of new information technologies reach everyone in a safe and fair way.
The proposed method combines a graph neural network with hierarchical clustering to extract key features linking crystal structure to optical spectra, and then groups materials with similar structural and spectral characteristics, revealing patterns that can guide materials design.
A method to interpret artificial intelligence (AI) models used in materials discovery by analyzing their learned features has been developed by researchers from Japan. The method extracts key features from an AI model trained on atomic structural data and optical absorption spectra, and then groups materials with similar structural and spectral characteristics. This approach can be extended to reveal how atomic arrangements influence other material properties, paving the way for more efficient materials design.
In recent years, artificial intelligence (AI) has emerged as a powerful tool to predict how materials will behave based on their atomic structure, helping researchers discover new materials faster and reduce reliance on trial-and-error methods. However, many of these models work like “black boxes.” They can make accurate predictions, but they do not explain how those predictions are made. This makes it difficult to understand the relationships between a material’s structure and its properties, limiting how useful these models are for guiding the development of new designs.
Now, in a study to be published in the journal Advanced Intelligent Discovery on June 15, 2026, researchers from Institute of Science Tokyo (Science Tokyo), Japan, have developed a method to make these models more interpretable. Their approach works by analyzing a trained AI model and extracting the key features it has learned about how crystal structure relates to optical spectra. Using these features, the researchers then grouped materials that share similar optical spectra and structural characteristics.
The study was led by Assistant Professor Akira Takahashi, Professor Fumiyasu Oba (also a project leader at KISTEC, Japan), and Master’s course student Arata Takamatsu (at the time of the research) of the Materials and Structures Laboratory, Science Tokyo, in collaboration with Professor Yu Kumagai of the Institute for Materials Research, Tohoku University, Japan.
“Our proposed classification method allows for an understanding in detail of how AI prediction models make predictions, namely, extracting key factors for desired spectral shapes, and thereby providing useful physical and chemical insights for materials design,” says Takahashi.
Material’s properties often depend on some parameters and are described using spectral data—for example, optical absorption spectra capturing how light interacts with the material across different wavelengths. Compared to properties represented by a single number, spectral data are far richer and more complex, making them challenging to interpret using conventional AI methods.
The researchers used an atomistic line graph neural network (ALIGNN), an existing graph neural network architecture, trained to predict optical absorption spectra from atomic structure using data from 2,681 metal oxides, chalcogenides, and related compounds. From the trained model, they extracted features from its internal layers and applied hierarchical clustering, a method that groups items based on similarity. This allowed them to classify materials into distinct groups that shared both structural features, such as elemental composition, atomic coordination, bond lengths, and bond angles, and similar spectral shapes.
Notably, the model learned these patterns from atomic structure alone, without being given oxidation states or electronic configurations as input, indicating that it had internally captured meaningful relationships between structure and properties.
Optical properties play a key role in many applications. They affect a material’s appearance, which is important for pigments and dyes, and govern how it interacts with light in devices such as solar cells and photodetectors. Understanding which elemental species and structural features shape these spectra is, therefore, key to establishing rational design guidelines for such materials.
Furthermore, the approach is not limited to optical spectra: it can be extended to determine how a material’s structure influences its behavior under different conditions, such as temperature or pressure, opening up new possibilities for designing materials with specific and useful properties. As demonstrated here for optical absorption, the approach can be applied to a range of spectral properties, enabling researchers to identify common factors shared by different materials and infer the origins of desired spectral characteristics.
“It has been difficult to interpret what machine learning models have learned about spectral properties. In this work, we developed a general method to extract such insights, which we believe will prove broadly useful for materials research,” concludes Takahashi.
***
About Institute of Science Tokyo (Science Tokyo)
Institute of Science Tokyo (Science Tokyo) was established on October 1, 2024, following the merger between Tokyo Medical and Dental University (TMDU) and Tokyo Institute of Technology (Tokyo Tech), with the mission of “Advancing science and human wellbeing to create value for and with society.”
Deep Learning-Based Extraction of Promising Material Groups and Common Features from High-Dimensional Data: A Case of Optical Spectra of Inorganic Crystals
Article Publication Date
15-Jun-2026
COI Statement
The authors declare no conflicts of interest.
Rockefeller University Press partners with Cashmere for responsible AI discoverability
Partnership uses Cashmere’s infrastructure suite to enable safe, controlled and transparently reported AI inference use of Rockefeller University Press journals
Rockefeller University Press (RUP) has partnered with Cashmere, a data infrastructure platform, to manage the integration of its scientific literature into AI-powered research applications. The collaboration establishes a secure and transparent framework for AI inference, which is the phase in which a trained AI model queries live data to answer user questions and generate real-time results. The partnership ensures that RUP’s peer-reviewed content is accurately represented and discoverable to scientists who are using AI tools while simultaneously protecting that same content from unauthorized training use.
Through the partnership, RUP will make Journal of Cell Biology (JCB), Journal of Experimental Medicine (JEM), Journal of General Physiology (JGP), and Journal of Human Immunity (JHI) available for AI inference use cases. The arrangement ensures that AI applications can access RUP content within clearly defined licensing boundaries. RUP retains full visibility into how its content is used, by whom, and to what commercial effect, with detailed usage analytics and transparent revenue reporting delivered through the Cashmere platform.
“Life sciences is one of the domains where the quality and provenance of AI’s source material matters most," says Jonathan Munk, CEO, Cashmere. "When a clinician, researcher, or AI agent draws on a journal article to inform a decision, they need to know that content is peer-reviewed, current, and legitimately licensed. Rockefeller University Press has been producing exactly that kind of authoritative science for generations. Our partnership makes that content available to AI developers on terms that are clear, enforceable, and fair, and gives RUP the visibility and revenue transparency they deserve as a content partner in the inference economy.”
The partnership builds on Cashmere’s growing network of leading publishers and AI platforms, and it reflects broader momentum across the scholarly publishing sector to establish principled frameworks for AI content licensing.
“Rockefeller University Press has been committed to the widest possible dissemination of rigorous science, and AI represents a new channel through which researchers, clinicians, and developers engage with biomedical knowledge," says Rob O’Donnell, Senior Director of Publishing, Rockefeller University Press. "This partnership with Cashmere allows us to participate in that channel with confidence. We can view how our content reaches AI applications through a framework that respects our authors’ work, enforces appropriate licensing, and gives us genuine insight into how our publications are being used and the value they are generating. It is the kind of controlled, transparent approach that scholarly publishing needs to embrace the AI era responsibly.”
About Rockefeller University Press
Rockefeller University Press is committed to quality and integrity in scientific publishing. RUP publishes Journal of Cell Biology (JCB), Journal of Experimental Medicine (JEM) and Journal of General Physiology (JGP), and co-owns Journal of Human Immunity (JHI) and Life Science Alliance (LSA). We use the latest technologies and carry out rigorous peer review, applying the highest standards of novelty, mechanistic insight, data integrity, and general interest to fulfill our mission of publishing excellent science. RUP’s journals were established by the research community, and editorial decisions and policies continue to be driven by scientists who actively contribute to their fields, appreciate the value of peer review, and desire a better publication experience for all. Learn more at rupress.org.
About Cashmere
Cashmere is a data infrastructure platform that connects premium content publishers with AI companies. Built on its proprietary OmniPub infrastructure, Cashmere enables publishers to protect, license, and monitor their intellectual property while unlocking new revenue streams in the AI economy. Cashmere’s $5M seed round was led by Reach Capital, with participation from a group of strategic industry partners. For more information, visit www.cashmere.io. Contact press@cashmere.io.
New AI heart failure screening tool could save lives, cut hospital waiting lists and deliver major NHS cost savings
Scientists at Queen Mary University of London and Barts Health NHS Trust have invented an AI algorithm to screen underlying heart failure in a GP clinic or even at home. The test could dramatically speed up diagnostics and free up precious hospital resources.
The team have been awarded £1.3m by the National Institute for Health and Care Research to develop the technology for use in Britain’s National Health Service to bring down waiting times.
Over a million people in Britain have heart failure. East London, where both Queen Mary and Barts Health are based, has among the highest rates of heart and circulatory disease.
Scientists at Queen Mary University of London and Barts Health NHS Trust have invented an AI algorithm which could transform how heart failure is screened and diagnosed.
Known as iHeF (Intelligent Heart Evaluation Framework), the invention could save the NHS substantial sums each year by screening out patients who don't need scans. The technology could free up hospital diagnostic capacity, cut waiting times, and save lives by fast-tracking patients who need treatment.
The Queen Mary and Barts Health team have been awarded £1.3m by the National Institute for Health and Care Research (NIHR) Invention for Innovation (i4i) programme to develop the invention for use in the NHS to bring down waiting times.
The money will fund the testing needed to secure regulatory approval. The team will test iHeF against today’s gold standard approaches; work with patients, clinicians and technicians to optimise the design; and model what the overall health and economic impact could be.
Queen Mary and the Life Sciences team at Barts Health will run a real‑world evaluation of how iHeF fits into NHS care, working with clinicians and patients to assess how clinicians use AI‑generated insights, how patients respond to AI‑supported diagnosis, and how well the tool fits existing NHS workflows.
The project is expected to last three years, so iHeF could be in clinical use by the end of 2028.
Dr Nay Aung, Clinical Senior Lecturer at Queen Mary University of London, Consultant Cardiologist at Barts Heart Centre, and the lead inventor of iHeF, said:
“Over a million people in Britain have heart failure and an estimated 400,000 more remain undiagnosed. Early diagnosis saves lives and improves outcomes. Our invention aims to reduce waiting times so patients can be screened and diagnosed sooner and create substantial efficiency savings for healthcare providers.”
Diagnosing heart failure is expensive and slow
Over a million people in the UK suffer from heart failure, yet diagnosing it is expensive and slow. Doctors typically use an ultrasound test known as echocardiography, a complex process which requires specialised staff and can only be done in a hospital or specialised diagnostic centre. The NHS performed over 1.8m echocardiograms in 2024/25, but the waiting list remains stubbornly high at around 150k patients. One in four waits over six weeks, one in ten waits over three months.
At the same time, only 25-30% of these echocardiograms produce clinically actionable findings. Many patients tested either don’t have heart failure or have minor cardiac abnormalities which do not necessarily need medical attention. Tests that turn out negative cost hundreds of millions annually, one analysis estimated total pathway costs at around £600m, and delay those who do have heart failure from getting the attention they need.
The problem is likely to get worse as the population ages, driving up demand for diagnostics.
iHeF will make screening cheaper, faster, earlier, and more precise. iHeF can screen for heart failure from a routine Electrocardiogram (ECG) recording by detecting changes in the heart’s structure and pumping functions. ECG is far cheaper than an echocardiogram and is already done early on by clinical staff with minimal training, so the invention would not add any new steps to the current care pathway. The equipment is easy to use, so it doesn’t rely on highly specialised staff. Crucially, it can be done outside of a hospital, such as in GP clinics, Community Diagnostic Centres, or even in the home.
Colin Heath, a patient at Barts Health, said:
“I believe using AI to help check heart tests could really cut waiting times, making care quicker and less stressful for people like me. I also think that research like this, using the right data, can make a real difference and may even help save more lives in the future.”
Light-driven memory could slash AI energy use
Laser-written memory runs 1,000× faster, slashing energy use for AI and data centers
The National Institutes for Quantum Science and Technology
New artificial ferrimagnetic materials made of antiferromagnetically coupled Co/Gd/CoFeB trilayers enable reliable switching of magnetic states using a single femtosecond laser pulse. Developed with synchrotron analysis at NanoTerasu, these materials are compatible with magnetic tunnel junctions and could serve as key components in future high-speed, low-power memory and computing devices.
Credit: National Institutes for Quantum Science and Technology
As artificial intelligence, cloud computing and digital services continue to expand, the world is facing a growing need for faster and more energy-efficient ways to store and process information. A team led by the National Institutes for Quantum Science and Technology (QST) has now developed a new magnetic memory material that can be rewritten using laser light instead of electric current, a step that could help reduce power consumption in data centers and support future high-speed information systems. This study was published online in Applied Physics Letters on June 8, 2026 (DOI: 10.1063/5.0328535).
The new material allows magnetic information to be switched by a single ultrashort laser pulse. Because light can reverse magnetic states much faster than electric current, the approach could deliver switching speeds roughly 1,000 times higher than those of conventional electrically driven magnetic memory, while also reducing heat generation and energy loss. The researchers say the advance points to a new class of low-power magnetic memory for AI hardware, edge devices and future optoelectronic platforms.
“Today’s digital society needs memory technologies that are both faster and more sustainable,” said Dr. Seiji Sakai, Group Leader at the Quantum Materials and Applications Research Center, Takasaki Institute for Advanced Quantum Science, QST. “By showing that a practical memory material can be switched using light, we believe this work opens a realistic path toward ultrafast, low-power devices for future information systems.”
Magnetic memory stores information by changing the direction of magnetization inside a material. Existing magnetic memory technologies typically use electric current to write data. That approach is attractive because it can retain information even when power is turned off, but it also faces major limitations: writing speed is constrained, and current generates heat, which increases energy consumption. Those challenges are becoming more serious as AI and large-scale digital infrastructure continue to push power demand upward.
To address this problem, the team focused on all-optical switching, a phenomenon in which light reverses magnetic orientation without the need for current. This effect had previously been observed in ferrimagnetic materials, but those materials were not suitable for practical memory because their magnetic readout properties were too weak for stable digital operation. By contrast, the alloy CoFeB is already widely used in magnetic memory because it offers nearly complete spin polarization and excellent readout performance, yet it had not been considered suitable for optical switching.
The researchers overcame that barrier by designing a new artificial ferrimagnet built from antiferromagnetically exchange-coupled layers of cobalt, gadolinium and CoFeB. By tuning the thickness of each layer with atomic-scale precision and optimizing the full multilayer structure, they created a material in which magnetic states can be reversed reproducibly with a single femtosecond laser pulse. The team also showed that the write-and-rewrite operation could be repeated stably, demonstrating the basic functionality required for memory applications.
“One of the most important aspects of this work is that we achieved optical switching in a CoFeB-based system, which is already highly compatible with magnetic tunnel junction technology,” Sakai said. “That compatibility makes this result much more relevant for future devices than earlier demonstrations limited to model materials.”
A key part of the study was the use of NanoTerasu, Japan’s fourth-generation synchrotron radiation facility, where the team analyzed spin arrangements and interlayer interactions in the material using X-ray magnetic circular dichroism spectroscopy. These measurements provided atomic-level insight into the multilayer structure and played an essential role in guiding the design of the new material. The submission materials note that the ferrimagnets were developed using synchrotron analysis at NanoTerasu.
The potential impact of the work extends beyond laboratory demonstration. Faster, lower-power memory could help tackle one of the major hidden costs of the AI era: the rapidly increasing electricity demand of data centers and advanced computing systems. According to the submission materials, the technology may also serve in the longer term as a photoelectric conversion interface linking optical interconnects and electronic circuits, eventually contributing to integrated chips that combine photonics and electronics on the same platform. Practical use of such materials in optoelectronic interfaces could begin within the next decade.
About National Institutes for Quantum Science and Technology, Japan The National Institutes for Quantum Science and Technology (QST) was established in April 2016 to promote quantum science and technology in a comprehensive and integrated manner. The new organization was formed from the merger of the National Institute of Radiological Sciences (NIRS) with certain operations that were previously undertaken by the Japan Atomic Energy Agency (JAEA).
QST is committed to advancing quantum science and technology, creating world-leading research and development platforms, and exploring new fields, thereby achieving significant academic, social, and economic impacts.
About Dr. Seiji Sakai Dr. Seiji Sakai is a researcher at the Quantum Materials and Applications Research Center of QST. He focuses his research on magnetic and spin-electronic low-dimensional materials and their device applications. He has published over 130 papers on these topics, which have received more than 1,700 citations. Dr. Sakai is currently working on developing integrated spintronics and nanophotonics technologies to advance optoelectronic integration for next-generation information systems.
Mental health conditions such as depression and anxiety disorders affect millions of people worldwide, yet the underlying mechanisms are often difficult to study. There is a lack of model systems that adequately capture the complexity of human cognitive and affective processes, such as language or reasoning. An interdisciplinary research team from medicine, psychology, and computer science at the Else Kröner Fresenius Center (EKFZ) for Digital Health at TU Dresden has now shown that large language models (LLMs) can reproduce patterns of human emotions such as anxiety, sadness, or stress. In addition, they exhibit cognitive biases and can be specifically regulated using mindfulness-based strategies. This could establish AI models as a new, complementary method for basic psychological and psychotherapy research. The results were published in The Lancet Digital Health.
In their modelling study, the researchers investigated six large language models (LLMs), including GPT-4o and several Llama versions. They used standardized text prompts to induce seven different affective states in the models: anxiety, fear, anger, disgust, sadness, worry, and stress. These represent emotional states that also play an important role in many mental health conditions. The researchers then assessed the models’ responses using structured rating scales, as commonly applied in psychological research. In a next step, the team showed that these states can be reduced again through mindfulness-based emotion regulation strategies, such as a breathing exercise. In addition, they found that the models exhibited typical cognitive biases, such as a tendency to complete sentences more negatively after sadness had been induced. These biases in language output are well known from depression research in humans. The results suggest that LLMs can model certain cognitive and affective processes of humans in a simplified form.
Potential for research and new approaches in psychotherapy
LLMs could therefore be used in the future as a scalable and controlled model system in research – for example, to investigate specific mechanisms of mental health conditions or to test new therapeutic approaches in silico. Particularly in early stages of research, such models could help refine hypotheses and design studies more efficiently.
“Our results show that large language models can reproduce patterns of human affective and cognitive processes under controlled conditions. For psychology, this opens up the possibility to test hypotheses in a scalable and experimentally controllable system. We can use these models as tools to better understand underlying mechanisms and to explore new approaches – for example in talk-based psychotherapy,” says Dr. Magdalena Wekenborg, biopsychologist and head of the PsychoDigital research group at the EKFZ for Digital Health at TU Dresden.
Limitations of the study and open questions
The authors also emphasize the limitations of this approach: LLMs do not possess real emotions; their responses are based on learned patterns from training data. The intention is not to anthropomorphize artificial intelligence. Open questions therefore include the extent to which the findings can be transferred to human behavior, as well as the underlying mechanisms within the language models and their explainability. The researchers do not see LLMs as a replacement, but rather as a meaningful addition to psychological studies.
“One advantage of experiments with large language models is their reproducibility and scalability: we can repeat identical conditions as often as needed and systematically vary them. This enables new, data-driven experiments in psychological and biomedical research that were previously not possible,” says Prof. Jakob N. Kather, Professor of Clinical Artificial Intelligence at TU Dresden and physician at the University Hospital Dresden.
The publication highlights how collaboration at the interface of psychology, medicine, and computer science can open up new research approaches and reflects the core interdisciplinary strength of the EKFZ for Digital Health at TU Dresden.
Publication
Magdalena K. Wekenborg, Elizabeth A.M. Michels, Georg Kurze, Matti L. Kropp, Fabian Wolf, Josi Harzbecker, Isabella C. Wiest, Jakob N. Kather: Large language models as models of human psychopathology: a modelling study. The Lancet Digital Health, 2026.
Else Kröner Fresenius Center (EKFZ) for Digital Health
The EKFZ for Digital Health at the Faculty of Medicine at TUD Dresden University of Technology and University Hospital Carl Gustav Carus Dresden was established in September 2019. It receives funding of around 40 million euros from the Else Kröner Fresenius Foundation for a period of ten years. The center focuses its research activities on innovative, medical and digital technologies at the direct interface with patients. The aim here is to fully exploit the potential of digitalization in medicine to significantly and sustainably improve healthcare, medical research and clinical practice.Learn more: https://digitalhealth.tu-dresden.de/
