Anthropic's Claude Sonnet 5, the mid-tier model the company built to make AI agents cheap enough for everyday automation, just got 50% more expensive.
South Korean AI company Wrtn Technologies has closed a Series C funding round of roughly 100 billion won, betting that consumer AI entertainment scales as fast as enterprise tooling.
The average company running Salesforce's Agentforce platform now operates 13 distinct AI agents, up from five in early 2025 — with one large caveat about who was counted.
A $19.4 billion Microsoft-linked AI data center rising on a former baseball field in Vineland, New Jersey has become a test of whether permitting law can keep pace with AI's appetite for power.
Paris-based startup Uncovr has raised roughly €6 million in seed funding led by Index Ventures to attack one of medicine's most stubborn time sinks: the paperwork after the operation.
OpenAI's marquee bet on natural voice conversation has now had nearly two months in the wild, and the picture independent testing paints is more nuanced than the launch messaging.
OpenAI quietly closed the book on one of generative AI's founding products this weekend, folding a five-year-old brand into the chat window that replaced it.
A screenshot posted to X showed something the agentic-commerce industry has mostly only demoed: an AI agent taking an order all the way to checkout. Verifying it is another matter.
Alibaba's Qwen team closed out August by shipping Qwen3.8 Flash, an open-weight model that undercuts proprietary rivals on price while trading blows with them on vendor-run benchmarks.
As AI coding agents graduate from autocomplete tool to unsupervised committer, a new category of software is emerging — one that does not write code, but watches what the agents write.
A reasoning model that thinks for an hour drags a growing anchor behind it. A new preprint asks whether it needs to carry that weight at all.
Anthropic says it has automated a chunk of its own alignment research process — a result that is either encouraging or unsettling depending on which part of the field you sit in.
A new benchmark finds even the best commercial video-generation models solve barely half of a set of visual reasoning tasks — and rarely notice when they are wrong.
South Korea's Ministry of Science and ICT confirmed which companies will build a government-funded AI assistant for every resident — a test of AI as public utility.
X says it dismantled a network of roughly 200,000 accounts it links to China. Independent analysis suggests almost nobody saw the posts that made the story newsworthy.
Public Knowledge, one of the loudest critics of Congress's first attempt at a federal AI safety framework, has endorsed the second — and the reason why is the preemption fine print.
Australia's federal government has thrown its weight behind a scheme designed to get local AI companies into enterprise procurement pipelines — without mandating that anyone buy them.
At TechBBQ in Copenhagen, the question underneath every AI panel was not which models Europe can build, but who owns the infrastructure those models run on.