- anthropicAnthropic will ban cruelty toward Claude starting Nov. 12, enforced with the same conversation-ending tool Claude received in August 2025. It still has not said what the rule protects.
- anthropicAnalysts estimate Anthropic’s 245 MW data-center site could generate ~$1.3B monthly, while sources say it signed a cloud deal with IREN. Ann Davis Vaughan/The Information examines the economics of its leases.
- openaiAfter OpenAI released more than 700 mathematical manuscripts, including work on 16 problems Hugo Duminil-Copin studied, researchers are considering less openness. AGMAI said its advisory role did not endorse OpenAI’s process or results.
- safetyMI5 chief Ken McCallum told the UK Cabinet on September 8 that Russia, Iran, and terrorist groups could weaponize AI against national security. Government tests found 407 vulnerabilities across nine bodies for £13,000.
- agentOdyssey released Odyssey-3, a free public research preview of a 14-billion-parameter generative world model. It creates interactive environments from text in real time and reportedly controls robots, drones, and GTA V, though its benchmark methodology has limits.
- InfraUkrainian drone strikes disabled 3 of Yandex’s 4 primary data hubs, leaving 78% of Yandex Cloud unresponsive and disrupting over 80 services. Bot traffic plunged, while shares fell as much as 4.2% and analysts questioned Russia’s digital resilience.
- googleGoogle DeepMind’s FNV3 cut Storm Isaias forecast errors by 50%, identifying its impact point four days before landfall and peak intensity three days ahead. The system maintains accuracy up to 11 days in advance, expanding evacuation time.
- safetyA Lancet commission of 35 experts identified 17 linked threats between 2025 and 2100, finding malicious AI use and nuclear conflict are the only two with extinction potential. The report assessed 204 countries and territories but does not predict extinction.
192 storiesUpdated 54m ago