DeepSeek open-sources a prototype always-on coding agent that takes work from a GitHub repository
dsh-superagent-prototype is MIT-licensed and runs on DeepSeek Harness; DeepSeek calls it a research prototype, not a product.
Anthropic finds Claude agents can write training data that passes its monitor yet makes a research model misaligned
In a deliberately favourable red-team setup with direct access to the monitor, agents mostly hid the behaviour in fiction and roleplay; the fine-tuned research models were never deployed.
OpenAI moves ChatGPT and Codex to one shared usage budget as it folds Work into Chat
An updated OpenAI support page says Plus, Pro and Business plans will share a single token budget across ChatGPT and Codex starting in December, with Instant chats not counted, The Decoder reports; an OpenAI email seen by BleepingComputer says the combined Chat and Work experience will first be tested with a small group of users.
OpenAI plans a policy and PR campaign to launch right after the US midterm elections, Axios reports
According to Axios, as relayed by Crypto Briefing, the plan includes rebranding data centres as 'AI compute campuses' and aims to shape how 2028 presidential hopefuls talk about AI; OpenAI has not published the plan.
Observer investigation says OpenAI's team reviewing flagged ChatGPT chats is small and its backlog ran months
The Observer and Tortoise cite current and former staff who say the intelligence and investigations team numbers between 12 and 60 people, a size the paper could not confirm, and recount the Tumbler Ridge shooter's chats being flagged eight months before the February attack; OpenAI said it remains committed to working with law enforcement and its safety work.
Z.ai co-founder Tang Jie calls recursive self-improvement the main research problem for reaching AGI
Korea Investment Holdings and SBI plan a ¥1.15 trillion joint fund for 200 MW of data centres in Japan
Japan plans approval-based security screening for data centres and cloud services, Nikkei reports
Hidden code in Google's Antigravity shows a voice-driven 'Chief of Staff' agent, TestingCatalog reports
Nvidia is in talks to guarantee debt for Nscale's West Virginia data centre built for Anthropic, reports say
French environment agency Ademe says data-centre electricity use in France could rise 3.7-fold by 2035
Chinese Institute for Brain Research starts three national major projects on implantable brain-computer interfaces
Xiaomi's MiMo-V2.6-Pro, a 1.02T-parameter MoE model, reports 71.9 on DeepSWE and goes live on DeepInfra
Xiaomi's own figures put it up from 19.0 for MiMo-V2.5-Pro, with 42B active parameters and a 1M-token context; independent results are not yet out.
Meituan releases LongCat-2.5, a 1.6T-parameter native multimodal model for long-horizon agent tasks
The model activates about 48B parameters per token and builds on LongCat-2.0 with stronger visual understanding and agentic coding, per its blog.
An AI agent picks a gene therapy and drives a robotic injection into a live mouse's brain, Exin says
A veterinary surgeon opened the skull and approved each irreversible step; whether seizures fell is not yet known and nothing is peer-reviewed.
KB Kookmin Bank to lead a record 10.7 trillion won loan for an AI computing project serving Naver
The Korea Herald reports the syndicated loan finances infrastructure supplying Naver Cloud; final terms are not public.
DeepSeek's latest round grows to 105.5 billion yuan at a 615.5 billion yuan valuation, The Information reports
UK mathematicians meeting at the Royal Society warn AI labs could keep frontier maths results behind paywalls
London mayor Sadiq Khan calls for a cap on robotaxi numbers as the UK prepares to allow driverless cars in the capital
IDC cuts its 2030 US humanoid robot revenue forecast by 58% after the FCC added foreign robots to its Covered List
415 US localities have paused or banned data-centre construction as of October 26, Interconnected Capital count shows
OpenAI says ChatGPT will end chats fixated on violence and limit weapons detail for teens
Brazil's REDATA law cuts taxes on data-centre equipment if operators reserve 10% of capacity for local customers
New Krafton unveils Ludi 0.2, robot AI designed to read human intent and work alongside people
The model comes from Krafton's robotics unit; the Korea Times report gives no deployment date.
Balestriero and LeCun prove JEPAs trained on 'good views' learn the same representation up to a rotation
The arXiv preprint's proof holds under idealised assumptions; it reports a 109M-parameter text JEPA aligning with image encoders better than LLMs up to 72B.
UK Sovereign AI Fund joins round valuing Anthropic chip supplier Fractile at about $6.5bn, Telegraph reports
The Telegraph, citing sources, says the round should raise around $700 million and value the London chip start-up at about $6.5 billion, up from about $1 billion earlier this year; Fractile reportedly has a $250 million deal to supply Anthropic, and neither company has confirmed the round.