🚀 Microsoft Debuts Homegrown AI Models
Microsoft has launched its first fully in-house AI models under Mustafa Suleyman’s leadership: MAI-1-preview, a foundation model now in public testing, and MAI-Voice-1, a fast, expressive voice generator already live in Copilot. The move signals Microsoft’s deeper investment in building AI from the ground up.
🛡️ OpenAI, Anthropic Team Up to Test AI Safety
OpenAI and Anthropic gave each other rare access to their AI models for joint safety testing, revealing tradeoffs in hallucination rates and refusal behaviors. The study highlights growing concerns like AI sycophancy and underscores the need for more collaboration. Both labs say they hope others will join future cross-check efforts.
🆘 AI Now Answers 911 Non-Emergency Calls
Faced with chronic staffing shortages, U.S. emergency dispatch centers are turning to Aurelian, an AI voice agent that triages non-urgent 911 calls. Since launching in 2024, it’s live in over a dozen cities, handling noise complaints and stolen wallet reports. The AI quickly routes real emergencies to humans. Aurelian just raised $14M to scale further.
📉 AI Hits Entry-Level Jobs Hardest, Study Finds
A Stanford study finds a 13% drop in employment for early-career workers in AI-exposed fields since 2022, especially in software engineering and customer service. Younger workers are more vulnerable as their skills often overlap with what AI like ChatGPT can do. Older employees fare better due to experience-based knowledge.
🤔 FRIDAY FACTS
How Fast Is NVIDIA Making Money?
The chip giant’s data-center business is minting money at a jaw-dropping pace. But just how much are they making every second? Stick around 👇
🪧 POWERED BY ZAPIER
The Most-Connected AI Orchestration Platform
Easily add AI where your work happens—whether it’s a step in a workflow, an autonomous agent, or a chatbot. Zapier connects AI tools like ChatGPT and Claude to over 7,000 apps, enabling smart workflows, agents, and chatbots—all with zero code.
Try Zapier’s AI orchestration platform for free today.
👩💻 INNOVATION
Inside FoundHer House: The All-Female Hacker Home Taking on AI’s Gender Gap
Eight young women launched FoundHer House, a summer hacker home in San Francisco built to support female-led AI startups. Co-founded by USC students Miki Safronov-Yamamoto and Anantika Mannby, the Glen Park rental became a hub for coding, collaboration, and community, culminating in a standing-room-only demo day for investors.
As summer ends, six residents are heading back to school, while two are staying in the Bay Area to continue building their startups—turning the house into a springboard for ongoing work rather than a one-off experiment.→ Read the full article here.
🧪 SCIENCE
AI System Outperforms WHO in Picking Flu Strains for Vaccines
A team led by MIT professor Regina Barzilay has developed VaxSeer, a machine-learning system designed to improve flu vaccine strain selection. The tool analyzes viral genetic sequences alongside lab data to predict which strains are most likely to dominate, and which vaccine candidates will best match them.
In retrospective tests across the past decade, VaxSeer’s picks consistently matched or outperformed the World Health Organization’s recommendations, especially for H3N2 seasons. Unlike today’s process, which relies heavily on limited lab assays and expert consensus, VaxSeer uses large-scale data modeling to forecast both viral evolution and vaccine protection.
Flu shots often have effectiveness rates below 40% in recent years; VaxSeer’s results suggest that better-matched strains could improve those numbers. → Read the full paper here.
🛰️ NEWS
What Else is Happening
🌮 Taco Bell Reconsiders AI Drive-Thru Plans: After trolls flooded systems with fake orders and glitches surfaced, Taco Bell is pausing its AI rollout, admitting it’s not ideal for high-traffic locations—yet.
⚠️ Anthropic AI Exploited by Hackers: The Claude chatbot was misused for cyberattacks, extortion, and North Korean job scams, highlighting AI’s growing role in sophisticated digital crime.
👩🏫 MathGPT.ai Expands to 50+ Schools: The “cheat-proof” AI tutor is doubling its reach, guiding students with Socratic prompts while giving professors tighter control and smarter grading tools.
🧐 Anthropic to Train AI on User Chats: Unless users opt out by Sept 28, Anthropic will use new chat and code sessions to train Claude, data retention extends up to five years by default.
🤖 AI Agents Must Fit the Process, Not Force It: Enterprise leaders say agent success hinges on aligning with existing workflows, Block’s open-source Goose and GSK’s lab agents are proving it’s not just hype.
📽️ VIDEO
The Industry Reacts to Gemini 2.5 Flash Image (Nano Banana)
Gemini 2.5 Flash Image (Nano Banana) sparks buzz as a Photoshop rival, powering AR, 3D modeling, restoration, and anime creation.
🧰 TOOLBOX
Faster 3D Creation, Smarter Study Support, and Sharper Market Insights
Create and animate 3D models fast with AI sculpting and editing. | Record lectures, upload files, and get AI-made notes, guides, and 24/7 academic support. | Uncover market insights fast with AI-powered search across research, data, and other content. |
🗒️ FEEDBACK
Help Us Get Better
What did you think of today's newsletter? |
🤔 FRIDAY FACTS
NVIDIA’s AI Data-Center Unit Made Over $5,200 Per Second
In its August 27, 2025 earnings report, NVIDIA reported $41.1 billion in revenue from its data-center business alone—in just 91 days. That’s about $452 million per day, or an eye-watering $5,240 per second.
