Desk
Broad artificial intelligence coverage.
Block, owner of Square, Cash App and Afterpay, reports in an OpenAI case study that Daybreak and related tools let it replay detection changes against seven months of telemetry in under an hour and propose patches in 18 minutes during red-team exercises, with humans validating results.
NVIDIA says UK organizations in healthcare, finance, local government and climate are moving from AI plans to deployment on its stack, and pairs that with a five-year, $250 million compute commitment for agentic safety and security research and an Earthshot Prize partnership.
NVIDIA says cities and partners across Europe are building physical AI agents and digital twins on its Cosmos, Nemotron, Metropolis and Omniverse tools. France’s Côte-d’Or department reports cutting document search from weeks to minutes and operating costs by 10% across 500+ public buildings.
Argon’s developer headline is extreme output length—up to 1M tokens in one response—at intro prices that undercut several frontier rivals until November 30.
Google DeepMind and Singapore’s GovTech fine-tuned a lightweight Gemma model on local scam conversations so phones can flag manipulative chats without sending message contents to Google or the government. Rollout into citizen apps will be phased and require consent.
OpenAI’s EU economic research finds Codex and ChatGPT Work account for more than half of output tokens among sampled ChatGPT Enterprise customers, and that agentic use among EU small businesses grew more than eightfold since May. Even leading EU firms lag frontier firms elsewhere.
Prosecutors say an OpenAI engineer used chatbots and internal tools to plan planting a bug in an executive conference room. Motive is still murky; the insider threat is not.
Meta’s Muse Wonder puts a frontier agent model on a public Model API at $4/$20 per million tokens, with Wonder Full gated behind paid Muse Code plans.
NVIDIA’s cuDF Spark skills turn UDF rewrites into an agent loop—and one production example cut operator time from 274.55 hours to 12.78 on a 600-billion-row job.
The Verge’s account of the OpenAI bug case adds DOJ detail: CCTV near the executive TV, a second device in a bag, and OpenAI saying its models helped trace the plant.
U.S. District Judge Edward Davila granted Apple expedited discovery into OpenAI hiring materials for former Apple employees who interviewed for hardware and robotics roles, while denying broader requests into OpenAI’s hardware research for now. The ruling is procedural, not a finding on Apple’s allegations.
Google DeepMind says Gemini 4 Argon’s cyber and science capabilities triggered next-generation safeguards under its Frontier Safety Framework, including activation probes, mid-stream cutoffs and a misalignment monitor. Its reported results come from Google-selected tests and await independent replication.