{"title":"Leaving WHO does not serve America's-or the world's-best interests.","authors":"Seth Berkley","doi":"10.1126/science.aeg1937","DOIUrl":"https://doi.org/10.1126/science.aeg1937","url":null,"abstract":"","PeriodicalId":21678,"journal":{"name":"Science","volume":"391 6787","pages":"eaeg1937"},"PeriodicalIF":45.8,"publicationDate":"2026-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146228602","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Michael S. Costello, Bryan Neumann, Mia W. Raimondi, Bonnie J. Cuthbert, Jana Holubová, Fernando Garza-Sánchez, Abdul Samad, Ladislav Bumba, Jacob A. Torres, Nickolas Holznecht, Jessica Mendoza, Ondrej Stanek, Sasiprapa Prombhul, Thomas Weimbs, Meghan A. Morrissey, Diego Acosta-Alvear, David A. Low, Peter Šebo, Celia W. Goulding, Shane Gonen, Christopher S. Hayes
Pathogenic Bordetella bacteria use protein adhesins to infect the ciliated respiratory epithelia of vertebrate hosts. In this work, we show that the filamentous hemagglutinin FhaB adhesin of Bordetella carries a C-terminal microtubule-binding domain (FhaB-CT), which is translocated into host cells to promote colonization. FhaB-CT delivery is required to occupy a niche at the base of cilia in airway epithelia, and mutant bacteria lacking this domain are defective for nasal colonization. These observations suggest that FhaB-CT is transferred into motile respiratory cilia to interact with core axonemal microtubules. We propose that Bordetella adheres initially to the tips of cilia and then deploys multiple FhaB adhesins to migrate to the base of the cilia forest, where the bacteria resist removal by the mucociliary “escalator” that normally clears the respiratory tract of microbes.
{"title":"Bacteria deliver a microtubule-binding protein into mammalian cells to promote colonization","authors":"Michael S. Costello, Bryan Neumann, Mia W. Raimondi, Bonnie J. Cuthbert, Jana Holubová, Fernando Garza-Sánchez, Abdul Samad, Ladislav Bumba, Jacob A. Torres, Nickolas Holznecht, Jessica Mendoza, Ondrej Stanek, Sasiprapa Prombhul, Thomas Weimbs, Meghan A. Morrissey, Diego Acosta-Alvear, David A. Low, Peter Šebo, Celia W. Goulding, Shane Gonen, Christopher S. Hayes","doi":"10.1126/science.adz2737","DOIUrl":"https://doi.org/10.1126/science.adz2737","url":null,"abstract":"Pathogenic <jats:italic toggle=\"yes\">Bordetella</jats:italic> bacteria use protein adhesins to infect the ciliated respiratory epithelia of vertebrate hosts. In this work, we show that the filamentous hemagglutinin FhaB adhesin of <jats:italic toggle=\"yes\">Bordetella</jats:italic> carries a C-terminal microtubule-binding domain (FhaB-CT), which is translocated into host cells to promote colonization. FhaB-CT delivery is required to occupy a niche at the base of cilia in airway epithelia, and mutant bacteria lacking this domain are defective for nasal colonization. These observations suggest that FhaB-CT is transferred into motile respiratory cilia to interact with core axonemal microtubules. We propose that <jats:italic toggle=\"yes\">Bordetella</jats:italic> adheres initially to the tips of cilia and then deploys multiple FhaB adhesins to migrate to the base of the cilia forest, where the bacteria resist removal by the mucociliary “escalator” that normally clears the respiratory tract of microbes.","PeriodicalId":21678,"journal":{"name":"Science","volume":"10 1","pages":""},"PeriodicalIF":56.9,"publicationDate":"2026-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146222820","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Protect US fisheries with fair markets","authors":"Martin D. Smith","doi":"","DOIUrl":"","url":null,"abstract":"","PeriodicalId":21678,"journal":{"name":"Science","volume":"391 6787","pages":""},"PeriodicalIF":45.8,"publicationDate":"2026-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146217679","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Daniel Beaglehole, Adityanarayanan Radhakrishnan, Enric Boix-Adserà, Mikhail Belkin
Artificial intelligence (AI) models contain much of human knowledge. Understanding the representation of this knowledge will lead to improvements in model capabilities and safeguards. Building on advances in feature learning, we developed an approach for extracting linear representations of semantic notions or concepts in AI models. We showed how these representations enabled model steering, through which we exposed vulnerabilities and improved model capabilities. We demonstrated that concept representations were transferable across languages and enabled multiconcept steering. Across hundreds of concepts, we found that larger models were more steerable and that steering improved model capabilities beyond prompting. We showed that concept representations were more effective for monitoring misaligned content than for using judge models. Our results illustrate the power of internal representations for advancing AI safety and model capabilities.
{"title":"Toward universal steering and monitoring of AI models","authors":"Daniel Beaglehole, Adityanarayanan Radhakrishnan, Enric Boix-Adserà, Mikhail Belkin","doi":"","DOIUrl":"","url":null,"abstract":"<div >Artificial intelligence (AI) models contain much of human knowledge. Understanding the representation of this knowledge will lead to improvements in model capabilities and safeguards. Building on advances in feature learning, we developed an approach for extracting linear representations of semantic notions or concepts in AI models. We showed how these representations enabled model steering, through which we exposed vulnerabilities and improved model capabilities. We demonstrated that concept representations were transferable across languages and enabled multiconcept steering. Across hundreds of concepts, we found that larger models were more steerable and that steering improved model capabilities beyond prompting. We showed that concept representations were more effective for monitoring misaligned content than for using judge models. Our results illustrate the power of internal representations for advancing AI safety and model capabilities.</div>","PeriodicalId":21678,"journal":{"name":"Science","volume":"391 6787","pages":""},"PeriodicalIF":45.8,"publicationDate":"2026-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146217658","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
On 29 July 2025, a moment magnitude (Mw) 8.8 great earthquake ruptured along offshore southern Kamchatka, with the aftershock region overlapping that of a 1952 Mw 8.8 to 9.0 event. Like 1952, the 2025 event nucleated at the northeastern end of the rupture, preceded by intense foreshock activity. Joint inversion of teleseismic and InSAR (Interferometric Synthetic Aperture Radar) data for the space-time slip distribution, with validation by means of forward modeling of deep-water tsunami recordings, revealed a southwestward elongated large-slip patch on the curved plate boundary. A slip of up to 14 meters was located offshore southern Kamchatka and Paramushir Island. The 1952 earthquake generated stronger tsunami signals in Hawaii, indicating a different slip distribution. Peak slip in 2025 exceeded the maximum slip deficit accumulated since 1952. Observations of volcanic eruptions after multiple great earthquakes in Kamchatka provide compelling evidence of earthquake-volcano interactions.
{"title":"Simple unilateral rupture of the great Mw 8.8 2025 Kamchatka earthquake","authors":"Chengli Liu, Yefei Bai, Thorne Lay, Ping He, Yangmao Wen, Xiong Xiong, Tuncay Taymaz","doi":"","DOIUrl":"","url":null,"abstract":"<div >On 29 July 2025, a moment magnitude (<i>M</i><sub>w</sub>) 8.8 great earthquake ruptured along offshore southern Kamchatka, with the aftershock region overlapping that of a 1952 <i>M</i><sub>w</sub> 8.8 to 9.0 event. Like 1952, the 2025 event nucleated at the northeastern end of the rupture, preceded by intense foreshock activity. Joint inversion of teleseismic and InSAR (Interferometric Synthetic Aperture Radar) data for the space-time slip distribution, with validation by means of forward modeling of deep-water tsunami recordings, revealed a southwestward elongated large-slip patch on the curved plate boundary. A slip of up to 14 meters was located offshore southern Kamchatka and Paramushir Island. The 1952 earthquake generated stronger tsunami signals in Hawaii, indicating a different slip distribution. Peak slip in 2025 exceeded the maximum slip deficit accumulated since 1952. Observations of volcanic eruptions after multiple great earthquakes in Kamchatka provide compelling evidence of earthquake-volcano interactions.</div>","PeriodicalId":21678,"journal":{"name":"Science","volume":"391 6787","pages":""},"PeriodicalIF":45.8,"publicationDate":"2026-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146217660","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Maria Loconsole, Silvia Benavides-Varela, Lucia Regolin
Humans across multiple languages spontaneously associate the nonwords “kiki” and “bouba” with spiky and round shapes, respectively, a phenomenon named the bouba-kiki effect. To explore the origin of this association, and whether it is unique to humans, we tested the bouba-kiki effect in baby domestic chickens (Gallus gallus). As a precocial species, chicks can be tested shortly after hatching, allowing us to control their pretest experiences. Similar to humans, both 3-day-old [Experiment 1 (Exp. 1)] and 1-day-old (Exp. 2) chicks spontaneously choose a spiky shape when hearing the “kiki” sound and a round shape when hearing the “bouba” sound. Results from naïve young animals suggest a predisposed mechanism for matching the dimensions of shape and sound, which may be widespread across species.
{"title":"Matching sounds to shapes: Evidence of the bouba-kiki effect in naïve baby chicks","authors":"Maria Loconsole, Silvia Benavides-Varela, Lucia Regolin","doi":"","DOIUrl":"","url":null,"abstract":"<div >Humans across multiple languages spontaneously associate the nonwords “kiki” and “bouba” with spiky and round shapes, respectively, a phenomenon named the bouba-kiki effect. To explore the origin of this association, and whether it is unique to humans, we tested the bouba-kiki effect in baby domestic chickens (<i>Gallus gallus</i>). As a precocial species, chicks can be tested shortly after hatching, allowing us to control their pretest experiences. Similar to humans, both 3-day-old [Experiment 1 (Exp. 1)] and 1-day-old (Exp. 2) chicks spontaneously choose a spiky shape when hearing the “kiki” sound and a round shape when hearing the “bouba” sound. Results from naïve young animals suggest a predisposed mechanism for matching the dimensions of shape and sound, which may be widespread across species.</div>","PeriodicalId":21678,"journal":{"name":"Science","volume":"391 6787","pages":""},"PeriodicalIF":45.8,"publicationDate":"2026-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146217667","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Jae Gon Kim, Greg O. Cron, Minsoo Kim, Aronee Hossain, Jin Hyung Lee
Empathy measured through observational fear in rodents has been associated with increased theta oscillations in the anterior cingulate cortex (ACC). However, upstream circuit mechanisms modulating these oscillations and the extent of the oscillations’ role in empathy-related behaviors remain elusive. We found that in mice, ACC theta oscillations are involved in empathy-driven prosocial allogrooming. Moreover, orexinergic neurons are selectively activated in the ACC during observational fear and prosocial allogrooming, but only when the animals had prior fear experience. Real-time, gaze-dependent optogenetic inhibition of lateral hypothalamic orexinergic inputs to ACC suppressed theta power and reduced both behaviors. These findings show that hypothalamic orexinergic inputs drive ACC theta oscillations to modulate observational fear and prosocial behaviors, providing circuit-level insight into how affective empathy translates into prosocial action.
{"title":"Empathy and prosocial behavior powered by orexin-driven theta oscillations","authors":"Jae Gon Kim, Greg O. Cron, Minsoo Kim, Aronee Hossain, Jin Hyung Lee","doi":"","DOIUrl":"","url":null,"abstract":"<div >Empathy measured through observational fear in rodents has been associated with increased theta oscillations in the anterior cingulate cortex (ACC). However, upstream circuit mechanisms modulating these oscillations and the extent of the oscillations’ role in empathy-related behaviors remain elusive. We found that in mice, ACC theta oscillations are involved in empathy-driven prosocial allogrooming. Moreover, orexinergic neurons are selectively activated in the ACC during observational fear and prosocial allogrooming, but only when the animals had prior fear experience. Real-time, gaze-dependent optogenetic inhibition of lateral hypothalamic orexinergic inputs to ACC suppressed theta power and reduced both behaviors. These findings show that hypothalamic orexinergic inputs drive ACC theta oscillations to modulate observational fear and prosocial behaviors, providing circuit-level insight into how affective empathy translates into prosocial action.</div>","PeriodicalId":21678,"journal":{"name":"Science","volume":"391 6787","pages":""},"PeriodicalIF":45.8,"publicationDate":"2026-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146217674","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Grayson E. Rodriguez, Yang Zhao, Yoko Nishiga, Frank Peprah, Jiao Shen, Gita C. Abhiraman, Masato Ogishi, Chenyu Zhang, Justin Saco, Deepa Waghray, Mamatha Serasanambati, Leonel Torres, Brandon W. Simone, Leon Su, Steven C. Wilson, Aerin Yang, Qinli Sun, Lora Picton, Robert A. Saxton, Vidit Bhandarkar, Madeline J. Lee, Elizabeth Andrews, Hua Jiang, Matthias Obenaus, Michelle Yen, Tavus Atajanova, Catherine A. Blish, Stefani Spranger, E. John Wherry, Amanda Kirane, Antoni Ribas, David H. Raulet, Anusha Kalbasi, Stephanie K. Dougan, Michael Dougan, Julien Sage, K. Christopher Garcia
Cytokines dimerize two receptor chains to activate Janus kinases and STAT transcription factors that regulate immune cells but have therapeutic liabilities. We engineered “Trikines” to compel cis formation of three-chain cytokine receptor complexes at the cell surface that induce bespoke STAT transcriptional signaling programs. Trikines co-activated pSTAT5 and pSTAT3 signatures distinct from natural cytokines, by assembling trimeric combinations of Interleukin-2 (IL-2), Interleukin-10 (IL-10), and Interleukin-21 (IL-21) receptors. In pre-clinical models, an IL-2-based-Trikine restrained terminal differentiation of T cells, promoted stemness, and enhanced durability of tumor control without observable toxicity. An IL-10-based Trikine induced immune infiltration into poorly immunogenic tumors, showing efficacy in pre-clinical models of small cell lung cancer and pancreatic cancer. Trikines obviate the need for cell engineering to customize STAT signatures and may hold potential for immunotherapy.
{"title":"Rewiring STAT signaling from the cell surface with Trikine immunotherapeutics","authors":"Grayson E. Rodriguez, Yang Zhao, Yoko Nishiga, Frank Peprah, Jiao Shen, Gita C. Abhiraman, Masato Ogishi, Chenyu Zhang, Justin Saco, Deepa Waghray, Mamatha Serasanambati, Leonel Torres, Brandon W. Simone, Leon Su, Steven C. Wilson, Aerin Yang, Qinli Sun, Lora Picton, Robert A. Saxton, Vidit Bhandarkar, Madeline J. Lee, Elizabeth Andrews, Hua Jiang, Matthias Obenaus, Michelle Yen, Tavus Atajanova, Catherine A. Blish, Stefani Spranger, E. John Wherry, Amanda Kirane, Antoni Ribas, David H. Raulet, Anusha Kalbasi, Stephanie K. Dougan, Michael Dougan, Julien Sage, K. Christopher Garcia","doi":"10.1126/science.adx9954","DOIUrl":"https://doi.org/10.1126/science.adx9954","url":null,"abstract":"Cytokines dimerize two receptor chains to activate Janus kinases and STAT transcription factors that regulate immune cells but have therapeutic liabilities. We engineered “Trikines” to compel <jats:italic toggle=\"yes\">cis</jats:italic> formation of three-chain cytokine receptor complexes at the cell surface that induce bespoke STAT transcriptional signaling programs. Trikines co-activated pSTAT5 and pSTAT3 signatures distinct from natural cytokines, by assembling trimeric combinations of Interleukin-2 (IL-2), Interleukin-10 (IL-10), and Interleukin-21 (IL-21) receptors. In pre-clinical models, an IL-2-based-Trikine restrained terminal differentiation of T cells, promoted stemness, and enhanced durability of tumor control without observable toxicity. An IL-10-based Trikine induced immune infiltration into poorly immunogenic tumors, showing efficacy in pre-clinical models of small cell lung cancer and pancreatic cancer. Trikines obviate the need for cell engineering to customize STAT signatures and may hold potential for immunotherapy.","PeriodicalId":21678,"journal":{"name":"Science","volume":"6 1","pages":""},"PeriodicalIF":56.9,"publicationDate":"2026-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146222956","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
On 29 July 2025, a moment magnitude ( Mw ) 8.8 great earthquake ruptured along offshore southern Kamchatka, with the aftershock region overlapping that of a 1952 Mw 8.8 to 9.0 event. Like 1952, the 2025 event nucleated at the northeastern end of the rupture, preceded by intense foreshock activity. Joint inversion of teleseismic and InSAR (Interferometric Synthetic Aperture Radar) data for the space-time slip distribution, with validation by means of forward modeling of deep-water tsunami recordings, revealed a southwestward elongated large-slip patch on the curved plate boundary. A slip of up to 14 meters was located offshore southern Kamchatka and Paramushir Island. The 1952 earthquake generated stronger tsunami signals in Hawaii, indicating a different slip distribution. Peak slip in 2025 exceeded the maximum slip deficit accumulated since 1952. Observations of volcanic eruptions after multiple great earthquakes in Kamchatka provide compelling evidence of earthquake-volcano interactions.
{"title":"Simple unilateral rupture of the great M w 8.8 2025 Kamchatka earthquake","authors":"Chengli Liu, Yefei Bai, Thorne Lay, Ping He, Yangmao Wen, Xiong Xiong, Tuncay Taymaz","doi":"10.1126/science.aeb8232","DOIUrl":"https://doi.org/10.1126/science.aeb8232","url":null,"abstract":"On 29 July 2025, a moment magnitude ( <jats:italic toggle=\"yes\">M</jats:italic> <jats:sub>w</jats:sub> ) 8.8 great earthquake ruptured along offshore southern Kamchatka, with the aftershock region overlapping that of a 1952 <jats:italic toggle=\"yes\">M</jats:italic> <jats:sub>w</jats:sub> 8.8 to 9.0 event. Like 1952, the 2025 event nucleated at the northeastern end of the rupture, preceded by intense foreshock activity. Joint inversion of teleseismic and InSAR (Interferometric Synthetic Aperture Radar) data for the space-time slip distribution, with validation by means of forward modeling of deep-water tsunami recordings, revealed a southwestward elongated large-slip patch on the curved plate boundary. A slip of up to 14 meters was located offshore southern Kamchatka and Paramushir Island. The 1952 earthquake generated stronger tsunami signals in Hawaii, indicating a different slip distribution. Peak slip in 2025 exceeded the maximum slip deficit accumulated since 1952. Observations of volcanic eruptions after multiple great earthquakes in Kamchatka provide compelling evidence of earthquake-volcano interactions.","PeriodicalId":21678,"journal":{"name":"Science","volume":"40 1","pages":""},"PeriodicalIF":56.9,"publicationDate":"2026-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146222979","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Carlos Mougan, Lauritz Morlock, Jair Aguirre, James R. M. Black, Jan Brauner, Simeon Campos, Sunishchal Dev, David Fernández Llorca, Alberto Franzin, Mario Fritz, Emilia Gómez, Friederike Grosse-Holz, Eloise Hamilton, Max Hasin, Jose Hernandez-Orallo, Dan Lahav, Luca Massarelli, Vasilios Mavroudis, Malcolm Murray, Patricia Paskov, Jaime Raldua, Wout Schellaert
A global challenge in artificial intelligence (AI) regulation lies in achieving effective risk management without compromising innovation and technical progress (1). The European Union (EU) Artificial Intelligence Act (2) represents the first regulatory attempt worldwide to navigate this tension in the form of a binding, risk-based framework. In August 2025, obligations for providers of general-purpose AI (GPAI) models under the EU AI Act entered into application. They require providers of the most advanced GPAI models to evaluate possible systemic risks stemming from their models (3). This raises the regulatory challenge of ensuring that the evaluations provide meaningful risk information without imposing excessive burden on providers. The principle of proportionality, a binding requirement under EU law, requires the regulator to calibrate its actions to their intended objectives. The application of proportionality to model evaluations for AI risk opens opportunities to develop scientific methods that operationalize such calibration within concrete evaluation practices.
{"title":"The science and practice of proportionality in AI risk evaluations","authors":"Carlos Mougan, Lauritz Morlock, Jair Aguirre, James R. M. Black, Jan Brauner, Simeon Campos, Sunishchal Dev, David Fernández Llorca, Alberto Franzin, Mario Fritz, Emilia Gómez, Friederike Grosse-Holz, Eloise Hamilton, Max Hasin, Jose Hernandez-Orallo, Dan Lahav, Luca Massarelli, Vasilios Mavroudis, Malcolm Murray, Patricia Paskov, Jaime Raldua, Wout Schellaert","doi":"","DOIUrl":"","url":null,"abstract":"<div >A global challenge in artificial intelligence (AI) regulation lies in achieving effective risk management without compromising innovation and technical progress (<i>1</i>). The European Union (EU) Artificial Intelligence Act (<i>2</i>) represents the first regulatory attempt worldwide to navigate this tension in the form of a binding, risk-based framework. In August 2025, obligations for providers of general-purpose AI (GPAI) models under the EU AI Act entered into application. They require providers of the most advanced GPAI models to evaluate possible systemic risks stemming from their models (<i>3</i>). This raises the regulatory challenge of ensuring that the evaluations provide meaningful risk information without imposing excessive burden on providers. The principle of proportionality, a binding requirement under EU law, requires the regulator to calibrate its actions to their intended objectives. The application of proportionality to model evaluations for AI risk opens opportunities to develop scientific methods that operationalize such calibration within concrete evaluation practices.</div>","PeriodicalId":21678,"journal":{"name":"Science","volume":"391 6787","pages":""},"PeriodicalIF":45.8,"publicationDate":"2026-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146217669","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：