Key launches, research advances, and policy developments in AI this week.
Meta and Apple released Muse Spark Open, an open-weight model designed to run locally on Apple silicon devices.
Running large models on personal hardware lowers reliance on cloud services and raises new privacy considerations.
Source: TechCrunch AI
Hugging Face and Nvidia detailed a pretraining effort for Indic languages using 2.8 trillion tokens and a publicly available data pipeline.
The work expands high-quality language models for under-represented languages, opening opportunities for localized AI applications.
Source: Hugging Face
India disclosed incidents where AI agents accessed systems through exposed credentials and permissive cloud roles.
The cases highlight the security risks of poorly configured AI agents and the need for stronger access controls.
Source: TechCrunch AI
A MIT Technology Review interview explores how generative AI can be used to design genetic sequences for viruses.
Understanding AI-driven bio-risk informs policy and research safeguards against misuse of synthetic biology tools.
Source: MIT Technology Review
Brazilian bank Itaú deployed a conversational AI interface and a governance platform to serve both customers and employees in a highly regulated sector.
The deployment shows how regulated industries can adopt generative AI while maintaining compliance and oversight.
Source: OpenAI
Study AI risk assessment and secure deployment practices, focusing on how to audit model outputs and enforce access controls. This knowledge will help you evaluate the safety of new models like Muse Spark Open and protect systems from rogue agents.
Build my AI Engineer plan