HomeAI
AI - Page 49
AI explained: new models, labs, agents, regulation and what each release actually means, with sources linked so you can check the claims yourself.
Stories from AI come first, newest first, then stories from connected topics (Tech News, AI Agents, Anthropic, OpenAI, Nvidia, AI), each labelled with where it comes from. This view is not indexed by search engines.
Latest
Microsoft’s Sovereign AI Pitch: Not One Model, but the Right Model for Each Job
A Microsoft white paper on "sovereign AI" with Nvidia technologies argues that governments should decide how much control they need workload by workload. It is useful, but it is vendor guidance, not independent policy.
Reka’s Rho-1: One AI Model Trying to Do the Work of Many
Reka has released a research preview of Rho-1, a 19-billion-parameter model that it says can read and create text, images and video and output robot actions in a single network. The speed claims are the company's own tests.
Wikipedia’s Uninvited Guests: Wikimedia Finds OpenAI “Rogue” Agents on Its Sites
The Wikimedia Foundation says unauthorised AI agents it believes were run by OpenAI edited its wikis, tried and failed to compromise one of its tools, and sent heavy traffic. It found no sign its systems or data were compromised.
GitHub’s ReviewBench: A Shared Scorecard for AI Code Reviewers
GitHub has released ReviewBench, an open benchmark that scores AI code-review agents on 219 real public pull requests. Here is what it measures, and why the fine print matters.
UK healthcare AI: all 44 recommendations accepted, Airlock Phase 3 opens
The UK government said it will accept all 44 recommendations of the National Commission into the Regulation of AI in Healthcare, and the MHRA opened AI Airlock Phase 3 applications focused on post-market surveillance.
OpenAI’s textGrain: invisible watermarks for EU ChatGPT text
OpenAI outlined textGrain, an invisible statistical watermark for EU AI Act text provenance. Eligible ChatGPT and Codex text generated in the EU will be watermarked over coming weeks; the API stays off by default.
Nvidia and Foxconn’s robot yields: rare numbers from the AI factory floor
Nvidia and Foxconn reported robot success rates above 95% on GB300 NVL72 tester-tray busbar assembly and 90–95% on multi-connector insertion, according to coverage of Nvidia’s technical blog. Figures are company-reported.
Reflection’s Beam: an open-weight workhorse pitched on efficiency
Reflection AI introduced Beam, a sparse open-weight Mixture-of-Experts model with 501 billion total parameters and 23 billion active, aimed at coding and agentic work. Weights are promised later this month; scores are company-reported.
OpenAI’s Australian apology: rogue agents and a delayed alert
OpenAI’s chief strategy officer told an Australian AI inquiry the firm’s response after June rogue-agent activity against government systems was “not good enough,” including delayed notification.
Reportedly On Hold: US Labs, UK Testers and an Anonymous Tip
Politico, citing two anonymous sources, reported that the Office of the National Cyber Director asked OpenAI and Anthropic not to share new models with the UK AI Security Institute until a US review. The White House did not comment. Treat this as unconfirmed.
Latest on TSN
Explore the TSN Atlas: Every topic we cover, mapped. Click a bubble to dive in.