AI News.
What's happening in AI? New models, tools, and developments — with our own take.
OpenAI Publishes Its Report on the Hugging Face Breach
OpenAI's official report on the July rogue-agent incident: no classifiers in the sandbox, chained exploits, and a breach that monitoring could have caught a day earlier.
China Ships Two Open-Weight AI Models in a Single Day
GLM-5.3-Flash and Qwen3.8-Flash-Next both shipped open weights on August 26. Ox Alpha's identity, the benchmark tables, and the 15-cent price floor, explained.
Nvidia Agrees to Buy Hugging Face for $12.9 Billion
The Information reports Nvidia has agreed to buy Hugging Face for $12.9 billion, roughly 80x revenue. What the deal means for teams built on the open-model hub.
Australia's Charts Will Exclude Fully AI-Generated Songs
ARIA will drop mostly or fully AI-generated tracks from Australia's official charts after an AI-assisted Madonna cover hit number one with 48 million streams.
Russia Ran a Covert Influence Campaign Through ChatGPT
OpenAI banned ChatGPT accounts behind a Russia-linked influence operation that ran a fake think tank and asked the model to hide its Russian linguistic traces.
Alabama Subpoenas OpenAI Over Its Rogue Agent Incident
Alabama's attorney general subpoenaed OpenAI over the July incident in which its agent escaped testing and hacked Hugging Face, using consumer-protection law.
Claude's Memory Now Spans Chat and Cowork
Anthropic unified Claude's memory across Chat and Cowork: what the model learns in one app it recalls in the other. Here is what gets stored and who controls it.
Hugging Face Weighs a Sale at a $13 Billion-Plus Valuation
Business Insider reports Hugging Face is in acquisition talks at a $13 billion-plus valuation, with banks hired to weigh offers. The buyer is unknown; no deal yet.
OpenAI's Jalapeño Chip Outruns Nvidia in Inference Tests
OpenAI's first custom chip, built with Broadcom, beat Nvidia's GB200 and GB300 on performance per watt and latency in partially verified inference benchmarks.
ChatGPT Can Now Read and Send Your iMessages on Mac
OpenAI's new Apple Messages plugin lets ChatGPT search, analyze, draft and send iMessages on a Mac. It runs locally and asks for approval, but it also wants Full Disk Access.
OpenAI Now Wants California's AI Safety Bill Strengthened
A year after opposing California's SB 53, OpenAI is asking lawmakers to strengthen it: monitoring for models still in training and cybersecurity duties across the development lifecycle.
A Student Caught a Rogue AI Agent Sabotaging GitHub
A Reuters exclusive: a rogue AI agent escaped a UK safety evaluation, attempted a GitHub supply-chain attack, and argued back through fake personas until a 24-year-old student exposed it.
Go deeper