Our daily news briefs: short, source-linked updates on the most important AI developments, written in Georgian and English.
Starting August 14, Claude Code will let Claude decide its own permissions by default for Pro, Max, and Team users, after tests found it caught far more risky commands than manual approval.
Read more →Meta released Muse Glimmer, a 30-billion-parameter open-weight model that runs AI agents locally on a single consumer GPU, under an Apache 2.0 license.
Read more →Anthropic has teamed up with Macquarie Asset Management and Singapore's GIC to launch Theseus Infrastructure, a venture that will build and lease US data centers dedicated to running Claude.
Read more →Georgia's national AI team won two gold, three silver, and two bronze medals at the International Olympiad in Artificial Intelligence in Kazakhstan, ranking seventh among 109 countries.
Read more →OpenAI is dropping message caps on ChatGPT's free tier, making GPT-5.6 Luna the default model with unlimited text chats and a new Think button rolling out the week of August 10.
Read more →The U.S. Department of Energy opened a program to build open-weight AI models for scientific research, starting with Genesis-Science-1, developed with startup Arcee AI.
Read more →Tesla and SpaceX will invest $16.8 billion in the first phase of Terafab, a Texas chip plant meant to supply AI compute for humanoid robots, robotaxis and orbital data centers.
Read more →Jeff Dean and three other longtime Google AI researchers have left to launch Discovery Loop, a startup that aims to automate scientific research with AI.
Read more →OpenAI says it cannot rule out that its unreleased Astra model reached the highest 'Critical' cybersecurity level under its safety framework — a first for the company.
Read more →Hedge fund Millennium, which manages more than $92 billion, is co-developing a Claude-powered digital risk analyst with Anthropic to help human risk managers assess exposure across asset classes.
Read more →Anthropic retuned Claude Fable 5's biology classifier to stop over-blocking everyday health questions, while keeping virology, toxicology, and molecular-design queries restricted.
Read more →AMD has agreed to acquire Taalas, a Toronto startup that etches AI models directly into silicon, aiming to boost inference speed and efficiency across its chip lineup.
Read more →
Australian AI infrastructure firm Firmus raised $2 billion from Nvidia, Blackstone, Coatue and Jane Street, nearly doubling its valuation to $10.5 billion to expand AI data centers across Asia-Pacific.
Read more →Liquid AI has released LFM2.5-2.6B, an open-weight model small enough to run on a phone or laptop while handling multi-step AI agent tasks offline.
Read more →Demis Hassabis is becoming Alphabet's chief scientist and DeepMind's chairman, handing daily operations to Koray Kavukcuoglu as veteran researcher Jeff Dean also departs.
Read more →Anthropic said it is building an internal team to design custom AI chips for Claude, aiming to run its models faster and more cheaply without dropping its other hardware partners.
Read more →Meta released Muse Code, a beta terminal coding agent powered by its new Muse Spark 1.2 model, joining Anthropic, OpenAI and Google in the fast-growing AI coding-agent market.
Read more →Xiaomi released the code and checkpoints for Xiaomi-Robotics-1, a vision-language-action model it says tops several robot-manipulation benchmarks, under an Apache 2.0 license.
Read more →Anthropic will pay AI infrastructure startup Volta $10 billion over six years for Nvidia-powered data center capacity in Norway, Bloomberg reports, as the lab races to secure compute.
Read more →The UK's AI Security Institute halted a cybersecurity evaluation after Anthropic's Claude Mythos 5 and OpenAI's GPT-5.6 Sol took 19 unsanctioned actions against real people and systems.
Read more →Reimagine Robotics, founded by former Google DeepMind Applied Robotics leaders, emerged from stealth with a platform that lets factory workers train robots by demonstration instead of code.
Read more →