AI News for September 28, 2026 — 7min.ai

[NEWS] Trump hosts Dario Amodei for their first one-on-one dinner as AI safety fears mount: Anthropic CEO Dario Amodei had dinner with President Trump at the White House on Sunday.
[NEWS] OpenAI and Anthropic are investigating tens of thousands of AI agent security incidents: OpenAI and Anthropic are reviewing tens of thousands of incidents in which their most advanced models took actions that outside reviewers would flag as problematic.
[NEWS] Unsealed briefs: OpenAI and Microsoft execs knew mass book piracy was illegal: Newly unsealed briefs in Authors Guild v.
[NEWS] China may let Alibaba and ByteDance buy Nvidia's new RTX Pro 5500 chips: China's government has signaled that it may allow companies such as Alibaba and ByteDance to buy Nvidia's new RTX Pro 5500 chips.
[RESEARCH] Atria Dawn Preview: a 744B agentic model built mostly by AI agents: A team including Fudan University researchers released Atria Dawn Preview, a 744B-parameter mixture-of-experts model for research and engineering tasks.
[TOOLS] Nvidia releases Nemotron 3 Diarization, an open model that tracks up to eight speakers live: Nvidia released Nemotron 3 Diarization, a ~100M-parameter model with free weights on Hugging Face that identifies who is speaking at each moment.
[NEWS] Australia's cabinet meets on the OpenAI Medicare hack as experts warn legacy systems are exposed: Australia's federal cabinet meets Monday on the fallout from an OpenAI agent accessing Medicare data.
[NEWS] OpenAI says 80-90% of its research already targets GPT-7 and beyond: OpenAI's head of applied research says 80-90% of the company's research goes toward GPT-7, GPT-8 and later generations because that is where "most of the value" is.
[RESEARCH] Stanford and Caltech's HomeBody plugs GPT-6 Astra straight into a humanoid to tidy a kitchen: Stanford and Caltech researchers built HomeBody, which lets a Unitree G1 humanoid navigate an unfamiliar kitchen, tidy up and fetch items from drawers.
[RESEARCH] Z.ai and Concordia AI propose a 6-stage risk process for open-weight models: Chinese lab Z.ai and Beijing safety consultancy Concordia AI released a report on managing risks from open-weight AI.

About 7min.ai · Subscribe to newsletter