A European Commission-mandated independent expert review reports that artificial intelligence is transforming Europe with breakthroughs in precision medicine and climate resilience, while the opacity of “black-box” systems and the spread of misinformation threaten public trust and risk widening inequalities, and reveals deep divides between well-funded private-sector innovation and under-resourced public research, stark geographic imbalances among Member States, and overlooked potential for AI in social-good and environmental initiatives.
ai-frontiers.org reports that Cloudflare found more than 50% of internet traffic is non-human, including a 1,700%+ increase in requests from AI agents, and that agent identification systems can tie actions to registered records of identity and principal, while model deployment cards measure the real-world impacts of AI agents.
The authors of “Cross-dataset transportability of pediatric chest X-ray deep learning across three countries” report an internal AUROC of 0.976 with 95.1% sensitivity, and external AUROCs of 0.798 and 0.742 with frozen-threshold sensitivities of 6.2% and 0%, from a protocol that separately tests discrimination, probability calibration, fixed operating-point transport, shortcut-associated signal, and limited-label recoverability for paediatric pneumonia classification across datasets from three countries.
The authors of “Robust Lightweight Deep Learning Models for Oral Cancer Screening” report that their optimised MobileViTv2 models achieved an average sensitivity of 83.2 ± 1.5% and specificity of 86.0 ± 0.8% on a held-out test set, with the best model reaching 87.4% sensitivity, 86.5% specificity, and a negative predictive value of 97.2%, and that directly optimising hybrid architectures for the edge strictly outperforms computationally heavy paradigms.
“Intelligent Cause Prioritisation? An Analysis of AI Policy Priorities and Governance in Africa” reports that Africa remains relatively overlooked in AI discussions because it is largely a consumer of frontier AI systems and many countries face pressing development challenges, and that governments across Africa are eager to embrace AI for economic transformation.
The researchers behind GPS-Bench introduce an evidence-grounded benchmark for governance policy simulation that links policies to relevant actors, actor actions and downstream impacts, using legislative records, lobbying disclosures, regulatory documents, corporate filings, economic data and other public evidence, and report that fine-tuning on the grounded record gives the strongest actor-level impact prediction, while decomposition adds mechanism.
George Akume, Secretary to the Government of the Federation, said a programme had recorded about 1.87 million registrations across all 774 local government areas and trained more than 125,000 Nigerians in cloud computing, artificial intelligence, cybersecurity and software engineering, that the National Digital Cloud Policy was targeting $750 million in private investment over the next 24 months, and framed the country’s ambition as moving from technology consumer to builder, producer and exporter of digital solutions and artificial intelligence.
Today’s lead: AI trials involve governments less often than trials in general.