Full Summary
This Friday, June 5th, the AI landscape is buzzing with major developments, from national security to consumer health, and a significant new executive order. Both CNBC and Engadget confirm OpenAI will comply with President Trump's executive order, requiring AI companies to allow the federal government to review their models 30 days before public release. This voluntary order, signed June 2nd, aims to ensure safety and could impact how quickly new AI technologies reach the public. OpenAI's head of countries, George Osborne, emphasized the role of democratic governments in AI's use. Meanwhile, Meta's Muse Spark AI model API for developers faces repeated delays, as reported by The Hindu and asatunews.co.id. While Meta AI Chief Alexandr Wang initially announced its release for April, a spokesperson now anticipates a launch "this month," with no set date. NewsBytes adds that Meta's future AI models, including Muse Spark, will focus heavily on consumer health, potentially integrating features into apps like Instagram and WhatsApp, raising privacy concerns. Wang leads the newly-formed Meta Superintelligence Labs, or MSL, which developed Muse Spark as its first model. Observer.com reveals Wang calls Muse Spark an "appetizer," with more competitive models "cooking," and that it was not open-sourced due to internal alerts about biological risks. In other news, New Zealand's National Cyber Security Centre now has direct access to Anthropic's Claude Mythos AI model, according to Let's Data Science. This advanced AI can find and exploit software vulnerabilities and is part of Anthropic's Project Glasswing, expanding to 150 organizations across 15 countries for cyber defense. Independent tests show Mythos succeeded in expert-level capture-the-flag challenges 73% of the time. NVIDIA has unveiled Nemotron 3 Ultra, which Memeburn calls America's most powerful open AI model, scoring 48 on the Artificial Analysis Intelligence Index. Foreignpolicyjournal.com adds this open-source model has 500 to 550 billion parameters and is designed for advanced reasoning, delivering five times faster inference. Apple will even rely on Google's fleet of Nvidia chips, specifically Blackwell B200 data center chips, to power its overhauled Siri in September, utilizing Nemotron 3 Ultra. Finally, China Daily reports Hong Kong has significantly upgraded its homegrown AI model, HKGAI V3, with tenfold token compression efficiency and nearly a hundredfold better uninterrupted agent runtime. HPCwire highlights Dnotitia Inc.'s DNA 3.0, a new family of enterprise-ready AI language models for custom needs, now on Hugging Face, designed to integrate with specific data and workflows. These rapid advancements mean consumers can expect more sophisticated AI integrations in their everyday apps, but also increased government oversight and potential delays in new tech reaching the market as safety and regulatory frameworks catch up.