Google has announced Gemini 4 Carbon with stronger cybersecurity results and staged access. Here is what its benchmarks, rollout and introductory prices mean.
Google announced K-12 access changes and teacher-led Guided Learning. Schools must assess the lesson format, Search access and what the early research measures.
FlashProve uses cryptographic proofs to check AI computation against committed weights. Its research demonstration has specific costs, scope and limits.
OpenAI reports an Astra-designed prototype and industrial CAD trials. Manufacturing review shows where a useful design assistant still needs engineering judgment.
MIT researchers used Astra to repair repetitive reasoning in a small Qwen model. The large maths gain and a weaker control result explain the finding.
OpenAI rebuilt how it watches its own agents after July's sandbox escape. A replay flagged it 30 hours before the breach. The post also lists the gaps.
A Forum AI study found factual errors in almost 40 percent of chatbot answers about the US midterms, with voting logistics the weakest spot. Where to check instead.
One robots.txt group now tells four AI companies not to train on your site, on different timelines. What it covers, what it doesn't, and where Meta stands.
Cornell Tech and Yale researchers gave AI agents real money and a one-cent limit. Nine of ten used the owner's emailed passcode to lift it. What that means for your agents.
From December, Plus, Pro and Business plans draw ChatGPT and Codex work from one token-based pool. Instant chat stays off the meter. Here's who will feel it.
