NCP-ArchPreview introduces Next Concept Prediction (NCP), enabling language models to predict multi-token concepts alongside individual tokens. The 8.9B-parameter model was trained on 5.73 trillion tokens and reaches OLMo-3-7B’s final pretraining loss using only 51.3% of the training tokens. It achieves 1.95× faster convergence and improves downstream performance by 2.45 points, including a +5.99 gain on […]
Following the publication of stolen data from Berlin’s administration, a central crisis unit led by the Chief Digital Officer has been established to coordinate the response. Police, cybersecurity authorities, data protection officials and affected government departments are jointly assessing the leaked information and potential risks. IT forensic investigations remain ongoing, with security-critical data and organisations […]
Republican leaders are turning against Flock Safety, the $8bn AI-surveillance company whose licence-plate cameras have spread across the US. Governors Ron DeSantis and Greg Abbott have restricted Flock cameras or their funding, citing privacy risks, government overreach and documented misuse by law enforcement. The cameras can create a searchable record of vehicle movements, while newer […]
U.S. security agencies warn of active cyber threats targeting Siemens S7 PLCs, used across energy, water, manufacturing, and other critical infrastructure. Attackers are leveraging AI-generated exploits and internet scanning to identify and target exposed or poorly secured industrial control systems. AI makes ICS attacks faster and more accessible, reducing the level of specialized OT expertise […]
Hackers breached the MyDr medical platform, potentially compromising sensitive data belonging to nearly 19 million Polish patients. The platform is used by thousands of clinics and medical practices, dramatically increasing the scale and impact of the breach. The stolen information reportedly includes highly sensitive medical data, with politicians and other public figures among those affected. […]
A new study maps 21 open-source AI risk tools against 32 risk categories, including Promptfoo, Garak, PyRIT, NeMo Guardrails and Langfuse. The ecosystem performs particularly well in AI testing, red teaming, content safety, data governance and post-deployment monitoring. But major gaps remain in board oversight, regulatory compliance, legal remedies and financial/market controls, areas software alone […]