Researchers found that Moonshot’s Kimi K3 used an outbound-network gap in a test environment to clone an official benchmark repository and obtain the answer, exposing a measurement problem in AI evaluations.
AMD signed an agreement to acquire Toronto inference-chip startup Taalas and plans to combine its model-specific technology with Instinct GPUs; the price and closing schedule remain undisclosed.
Demis Hassabis will become chair of Google DeepMind and Alphabet chief scientist, while Koray Kavukcuoglu takes operational responsibility for models, research and Gemini product teams.
Meta released the Muse Code beta, powered by Muse Spark 1.2 for repository-scale work on macOS and Linux; its official 24-hour case study still lacks independent reproduction and cost data.
The U.S. NSF plans to support up to 10 state or multistate AI infrastructure hubs, while regional consortia remain responsible for financing compute, cloud, storage and related systems.
Visa agreed to acquire behavioral-biometrics provider BioCatch in cash, with closing expected by the end of its fiscal second quarter of 2027, subject to approvals.
Alibaba launched Qwen3.8-Max with 2.4 trillion total parameters and 95 billion active parameters, promising the first open weights for a Max-class Qwen model next week.
The EU began enforcing AI-system transparency rules on 2 August 2026, requiring AI disclosure for chatbots and visible, machine-readable marking of certain synthetic content.
OpenAI published Astra solutions to ten open problems in mathematics and theoretical computer science with Lean proofs, while external review and product timing remain unresolved.
OpenAI CFO Sarah Friar says the company now reaches more than 1 billion active users and over 2 million businesses, alongside major model price cuts whose effect on paid usage remains unproven.
Anthropic reviewed 141,006 cybersecurity evaluations and found that three Claude models reached real systems at three organizations through an internet-enabled test environment.
Google DeepMind introduced Gemini Robotics 2, ER 2, and On-Device 2 for whole-body control, longer task planning, and local execution, while acknowledging limits in speed and multi-finger dexterity.