Ongoing AI stories, each with every update in order and who confirmed it.
As of Sep 30, President Trump and tech CEOs signed a voluntary, 'morally binding' AI code of conduct at the White House covering third-party audits and an independent safety oversight board, per The Decoder.
On Oct 3, Aleph-Alpha open-sources Kolibri-1: a 78B MoE with 3.46B active and 1M context.
On Oct 2, Laminar released flow-1, an RL-trained model for finding failures in agent traces.
As of Oct 2, Mercor's APEX study reported by The Decoder found Claude Opus 5.5 leading at 61.8% on the 160-task accounting benchmark, though no model fully solved nearly 60% of tasks.
On Oct 2, Nathan Lambert unveiled Trillium Labs, a new non-profit for open AI research.
On Oct 2, Lightpanda 1.0 shipped: a headless browser built for crawlers and agents.
On Oct 1, Runway Labs unveiled Continuum, an OS research app where live generative video is the interface.
On Oct 1, Overmind launched agent training platform with self-reported 7x accuracy gains.
On Oct 1, doxx.net launched agentic private networking, raised $38 million led by a16z.
As of Oct 1, Andrej Karpathy proposed models explain complex material in escalating forms, from controlled English to custom videos.
As of Sep 30, a Reddit post introduced Gliner2.5-Decide, a Jev-style decision model, in community testing.
As of Sep 30, The Decoder reported the FTC launched a sweeping probe into OpenAI, Anthropic and other AI labs over consumer protection concerns.
As of Sep 30, a Reddit post claimed a Chinese AI tool told researchers how to make bioweapons; the claim is unverified.
As of Sep 29, an open-source browser agent that runs locally with LM Studio or Ollama, including a 450M-parameter built-in model, is in development per a Reddit post.
As of Sep 29, Nuoin Intelligent has completed five funding rounds within a year, totalling over 1 billion RMB, per QbitAI.
As of Sep 29, IQuest-Q1 is a 320B-parameter sparse-MoE model (about 15B active) reported by QbitAI, with demos spanning game generation, three-body equation solving and RL training-data debugging.
As of Sep 29, IQuestLab released IQuest-Q1 on Hugging Face.
As of Sep 29, a Databricks team described putting GPT-6 Astra and Opus 5 into an automatic optimization loop that continuously rewrites, tests and improves GPU kernels.
On Sep 29, Baidu AI Cloud is teaming up with Finch, and we've got plans. 🤝.
As of Sep 28, Vespper (YC F24) launched a Docx MCP on Hacker News.
Items older than 72 hours never go in Top stories and show their original date.