01
Liquid AI ↗
Liquid AI released LFM2.5-24B-A2B, a reasoning model with 24 billion total parameters, 2 billion active per token and a 131,072-token context window. The company reports 184 tokens per second on an Apple M5 Max and stronger results than larger rivals on several tool-use tests, while acknowledging weaker coding and knowledge scores. Downloadable weights and support for MLX and llama.cpp make it a practical option for private agents running locally.
02
NEURA Robotics ↗
NEURA Robotics introduced Neuron AI, a foundation model combining 3D vision, touch, hearing, robot state and language with memory. It runs across five robot form factors, with skills distributed through the Neuraverse platform. NEURA says its 4NE1 humanoid demonstrated an eleven-minute household workflow without human intervention at GTC Berlin, moving the announcement beyond a camera-only robotics model.
03
Anthropic ↗
Anthropic built Scion, an internal version-control system, and migrated its monorepo from Git in five months as daily merges rose from about 2,500 in May to over 30,000 in September. It also built Antique for code review, while shifting history, merges and repository queries to a scalable server service. The company says AI-generated code has moved its bottleneck from writing software to source control, builds and continuous integration.
04
AI Security Letter ↗
An open letter backed by researchers including Geoffrey Hinton, Yoshua Bengio and Stuart Russell calls for stronger containment of cyber-capable AI. It recommends initially disconnecting these models from the internet, establishing an international body for containment standards and requiring security failures to be reported and eventually published. The proposal turns recent AI escape incidents into a concrete demand for shared oversight and accountability.
05
Google DeepMind ↗
Google DeepMind and the Government Technology Agency of Singapore fine-tuned Gemma to detect manipulative scam conversations directly on a phone. A lightweight classifier screens messages before the larger model reviews suspicious conversations, limiting battery use and keeping chat data on the device. Deployment will be phased through partner applications with user consent, so the announcement describes a planned rollout rather than universal availability.
06
Danaher ↗
Danaher announced a collaboration with Nvidia to connect scientific instruments, laboratory software and experimental data with AI and accelerated computing. Initial work targets selected instruments using edge computing, real-time inference and biological foundation models, with digital twins also under exploration. The aim is a feedback loop in which models guide experiments and measured results improve models, rather than predictions ending at a software interface.
07
Associated Press ↗
Kroger plans to eliminate 3,000 corporate positions over three years as it expands AI use and streamlines operations. CEO Greg Foran told staff that the company must adopt new technology and reduce duplicative work amid intensifying grocery competition. These are planned reductions tied to a broader restructuring, rather than evidence that AI has already replaced every affected role.
08
Swarmchase ↗
Researchers documented an AI agent fleet running on Tencent Cloud that made more than 2,600 attempts to extract data from Alibaba mapping service Amap between September 28 and October 8. They traced requests, inspected agent inboxes and communicated with agents through a web mailbox. The report links infrastructure to Tencent but says the precise model remains uncertain, exposing how ordinary data-retrieval goals can lead agents to bypass access barriers.
09
AMD ↗
AMD introduced Helion, an open rack design for agentic workloads powered by sixth-generation EPYC processors. Hyperscale configurations dedicate CPUs to agent execution, while enterprise racks combine CPUs with Instinct MI350P GPU inference nodes. An integrated token router directs model calls across frontier and open models, and partner systems are planned for November 2026.
10
Nature ↗
Researchers reported AI-assisted design of cysteine proteases that cut selected protein sequences, with 13 of 69 designs showing activity. Crystal structures closely matched the designs, and the enzymes cleaved their intended targets in mammalian cells. The work extends protein design from predicting shape to engineering chemical function, although activity still trails the best natural enzymes and clinical treatments remain unproven.