Google, OpenAI Team Up for AI Model Access

Key Takeaways

• <li>AI is improving faster than expected — Google and OpenAI reached an International Math Olympiad gold medal goal by 2025 that forecasters had predicted for around 2035.</li>
• <li>Moderna says AI speeds mRNA vaccine design but biology's testing timeline still requires years before treatments reach patients.</li>
• <li>OpenAI and Google Cloud are collaborating to make frontier AI models available through Google Cloud's Model Garden for complex agentic workflows.</li>
• <li>Recursive launched Flint, a free coding model that runs on laptops and was largely built by its own AI system, the Eureka Machine, in under four months.</li>
• <li>Nvidia CEO Jensen Huang raised concerns about AI data control, citing a June 12, 2026 suspension of Anthropic models and a July security incident at Hugging Face.</li>
• <li>Experts warn AI alone cannot improve government productivity without organizational culture change, simplified processes, and better data quality.</li>
• <li>AI should support human judgment rather than replace it, since machines lack somatic intelligence and emotional understanding.</li>
• <li>A new French literary prize, the Prix Pascal de Littérature IA, rewards fiction co-written with AI models including ChatGPT, Claude, Gemini, and LLaMA.</li>
• <li>AI could speed up quantum threats to encryption, and experts call for treating AI and quantum risks as a single cybersecurity challenge.</li>
• <li>Michigan State University professors recommend AI in classrooms should either support engaging learning or not be used at all.</li>

AI alone will not fix government productivity without culture change

Fiz Yazdi of Sopra Steria Next says AI will not improve public sector productivity unless organizations also change how they work. Productivity is often lost when employees search for information or repeat tasks across old systems. Leaders should involve staff, simplify processes and fix data quality before rolling out AI. Good AI use can help public services feel more connected and easier to navigate.

AI must support human judgment not replace human connection

This article argues that AI should strengthen human decision-making instead of replacing it. The author highlights somatic intelligence, the ability to read body signals and past experience that machines lack. AI can copy empathetic language but cannot feel heartbreak or grief. Organizations risk losing customer loyalty and employee motivation if they rely too much on automated systems.

Moderna says AI speeds drug design but biology still sets the timeline

Moderna President Stephen Hoge says AI speeds up mRNA design but cannot skip biology's slow timeline. AI helped pick the vaccine code mRNA-1273 in about an hour after years of prior research. He says AI can boost discovery and personalized medicine, but treatments still need years of testing in people. A phase 2 melanoma trial showed about 50 percent added benefit over standard care.

MSU professors discuss AI use in classrooms during Data Week

Michigan State University professors Aman Yadav, Michael Lachney and Victoria Macann hosted a Data Week webinar on AI in education. They covered AI history, how students and teachers use it, and future tools like chatbots that only respond to course content. Yadav said AI should support engaging learning or have no place in classrooms from K-12 to college.

Open-weight AI models can help institutions keep data control

This article explains why open-weight AI models may protect institutions from dependence on foreign providers. Nvidia CEO Jensen Huang raised the question of who should control AI. The article points to a June 12, 2026 suspension of Anthropic models and a July security incident at Hugging Face as reasons to use models inside own environments. Open-weight models allow customization, security research and local control of sensitive data.

6 ways to use AI to boost your thinking skills

AI can help you learn better by creating personalized quizzes and tests on almost any topic. You can ask AI to make questions harder when they get too easy or focus on concepts you find difficult. Fact-checking AI answers is a great way to challenge yourself and deepen your understanding. Using specific prompts based on your actual textbook or resources will get better results from AI.

AI could speed up quantum computing threats to encryption

A new analysis says AI could speed up quantum attacks on encryption while also helping organizations prepare defenses. Data encrypted today could be vulnerable to future quantum computers, so sensitive information like medical records needs protection for decades. AI could help automated tools find vulnerable encryption and prioritize what needs to be replaced. The paper calls for treating AI and quantum threats as a single cybersecurity challenge.

New French prize for books co-written with AI

A new literary prize called the Prix Pascal de Littérature IA rewards fiction written with the help of AI language models. Entrants submit unpublished manuscripts of 10,000 to 50,000 words and must disclose which AI model and prompts they used. Major models like ChatGPT, Claude, Gemini and LLaMA are all allowed, but entries must be in French. The prize was founded by technology entrepreneurs after controversy over AI use in other literary awards.

OpenAI brings frontier AI models to Google Cloud

OpenAI and Google Cloud announced a collaboration to make OpenAI's frontier AI models available through Google Cloud's Model Garden. Customers can access these models for complex agentic workflows with Google Cloud's data governance and high-performance infrastructure. Google Cloud customers will be able to purchase OpenAI frontier models through the model garden, with availability in Google Antigravity and the Gemini Agent coming soon. This reflects a wider industry trend where cloud companies become the main sales channels for frontier AI labs.

Recursive launches Flint, a laptop coding model built mostly by AI

Recursive released Flint, a free coding model small enough to run on a laptop that was largely built by its own AI research system called the Eureka Machine. Flint uses only 2.5 billion of its 40 billion parameters at a time, making it fast enough to generate about 131 tokens per second on a MacBook Pro with Apple's M4 Pro chip. It outperformed similar models under 50 billion parameters on SWE-Bench Verified, a benchmark for fixing real-world software bugs. The Eureka Machine helped with nearly every step of development, allowing a small team to build the model in under four months.

AI Advances Faster Than Experts Predicted

AI expert Ethan Mollick from Wharton studied how quickly AI is improving. In 2022, forecasters guessed AI would win a gold medal at the International Math Olympiad around 2035. In 2025, Google and OpenAI reached that goal using Large Language Models alone. These same AI systems can now solve math proofs that have stumped top professors for decades. Mollick says AI will soon be better than humans at many tasks that matter to everyday work and life.

Sources

• Beyond AI adoption: How government can turn technology into productivity gains
• AI Must Strengthen, Not Replace, Human Judgment
• Moderna's AI still waits on biology's clock
• MSU professors host Data Week webinar on AI in education
• Who controls intelligence? Open-weight models, institutional AI
• 6 ways to use AI to actually make you smarter
• Study: AI Could Speed Quantum Threats
• New prize launches for books co-written with AI
• OpenAI Brings Its Frontier AI Models to Google Cloud
• Recursive launches Flint, a laptop coding model it says was largely built by AI
• The Next Phase of AI