AI models, services and robotics
On September 17 Anthropic published the results and the code from Claude's optimization of 36 open-source biology model implementations. In just under four weeks structure prediction got 4.1x faster on average, and molecular machines the size of a ribosome were folded on a single GPU node.
On September 17 OpenAI released Astra for Law, its GPT-6 Astra model configured for legal work. It is not a new model but a configuration wrapped around a legal search index of more than 230 million URLs, drafting instructions, and controls built for law firms.
Nous Research opened a plugin catalog for Hermes Agent on September 16 with four official plugins and 96 from the community. Installs are pinned to commit hashes, capabilities require consent, a security scan runs at install time, and the documentation states plainly that being listed is not a code audit.
Toronto startup smartARM has shown a bionic arm prototype that uses a palm camera and Meta's open-source vision model DINOv2 to recognize objects and pick its own grip, in a video posted to Meta's newsroom on September 16. A few photos teach it a new object, and Meta AI glasses become a second pair of eyes.
On September 16 OpenAI announced a framework that sets when and how it will disclose cases of misalignment in its models, together with six incident reports from the past six months. Any employee can raise a case, each step has a deadline, and cases are sorted into three tracks. Behind it is the company's own judgment that alignment is not sufficiently solved.
On September 16, OpenAI released an admin guide that ties ChatGPT Work and Codex usage, tasks, and performance data to business value. The company says the analytics numbers are only a starting point — business leaders still have to work out the actual payoff.
On September 16 Anthropic folded Claude Cowork and chat into a single conversation and launched Claude Docs and Claude Slides. Deciding where a task belongs is now Claude's call, not yours.
On September 16, OpenAI started testing sponsored agents that let people chat with a company-sponsored agent after clicking a ChatGPT ad, running the trial with a limited group of US advertisers. The same day also brought the natural-language Ads Manager plugin for building campaigns, plus HubSpot and Shopify integrations.
OpenAI Academy partnered with AARP's OATS to hold free in-person ChatGPT workshops for 1,000 older adults across 10 U.S. cities on September 16. The share of ChatGPT messages linked to users 55 and older in the U.S. has grown from 6% to nearly 10% in a year.
The official ChatGPT account said on September 15 that GPT-5.5 will be pulled from every ChatGPT, ChatGPT Work, and Codex plan on October 14. OpenAI's developer account quickly drew a line, saying the model stays available through the API and Codex sessions authenticated with an API key.
The OpenAI Developers account posted a one-minute video of Cognition co-founder Walden Yan on September 15. Inside Devin, GPT-6 Astra backs up the claim that code works with tests, cutting the amount of code people have to read themselves.
Sakana AI added Interactive Reading and PowerPoint export to its research agent Sakana Marlin on September 16. Click a citation and the AI points to the supporting passage in the source, and research findings now come out as editable slides.
Google published an overview of two decades of language AI research on September 15 under the name of SVP James Manyika, covering technology that now reaches 300 languages and 7 billion people. Its backbone is models that process audio directly instead of translating text, and language data recorded by the communities themselves.
Google Research on September 15 published Retrieve-for-Train, a framework that returns search results as one coherent set, alongside its ICML 2026 paper. Running reinforcement learning once to teach a small diffusion model made query fan-out 12 to 20 times faster.
Anthropic's developer account posted a 2-minute-7-second founder interview video shot at Frontier Day in San Francisco on September 16. Four companies, Spawn, Hex, Sazabi and Phylo, describe what building a company on Claude looks like right now.
Google published a 26-page crisis resilience white paper on September 15 under the names of Yossi Matias, vice president of Google Research, and Kate Brandt, chief sustainability officer. It gathers the company's AI disaster-prediction work in one volume, from flood forecasts seven days out and cyclone forecasts 15 days out to cases where warnings turned into cash assistance and paused loan repayments.
Google released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on September 15. The two voice models run tools in the background mid-conversation and reason while they speak, and Extended Thinking took first place on the Speech to Speech Quality Index with a score of 82.6.
Meta opened Meta One, a subscription service bundling Instagram, Facebook, WhatsApp, and Meta AI, worldwide on September 15. The individual bundles raise AI image and video generation limits, while the creator and business tiers sell follow buttons and a business agent for a monthly fee.
Anthropic opened Salesforce in Claude as a public beta on September 15. The plugin pulls accounts and pipeline into Claude with 37 sales skills built in, and it is the first product of the Claudeforce partnership announced on August 26.
Bilal Chughtai, who worked on AGI safety and alignment research at Google DeepMind, warned of AI risk in a resignation post published on X and LinkedIn on September 14. He wrote that AI has the potential to kill us all, called for coordination and transparency to end the race between companies, and the post was read 560,000 times within a day.
The Tencent Hunyuan team released the EvolveScaler paper and project page on September 15, an information-evolution data framework. On data built by defining state in code first and then rendering it as natural language, the median avg@5 across 14 models fell to 11.3 on the longest tier.
On September 14 Nous Research announced a team tier for Hermes Agent, Hermes Business, and an on-premises edition, Hermes Enterprise. On top of a shared balance and per-member caps sits a skill library that the whole team uses, no matter who built each skill.
Microsoft Research featured a paper analyzing real-world GitHub Copilot coding-agent traces in its September 14 Research Focus. Of 761 million LLM calls, 87% were started by the agent itself, and a single tool failure could inflate compute by up to 4x.
Cohere CEO Aidan Gomez said live on Bloomberg TV on September 14 that the existential-risk debate around AI has veered too far into science fiction. He cited the Hugging Face hacking incident as an example of over-extrapolation and called it a red flag for the largest companies to set the rules among themselves.
