While you were drafting emails, AI Learned to Invent Matter

While the world remains fixated on chatbots, Scientists are quietly teaching AI not just to process our world, but to design a new one from the atoms up.

While you were drafting emails, AI Learned to Invent Matter
Daily AI BriefPublished
Share this AI Brief
Tags (8)
Generative AIAI in Material ScienceAI Cybersecurity RisksDeepfake TechnologyAI RegulationFuture of Work with AILarge Language ModelsDigital Trust

Unvritt.

AI Isn't Just Thinking Anymore. It's Building Our Reality.

Wed, Sep 24, 2025

AI learned to invent matter. No, that’s not the plot of a new sci-fi movie; it’s happening at MIT. While the world remains fixated on chatbots, the real revolution is happening in the background. Scientists are quietly teaching AI not just to process our world, but to design a new one from the atoms up.

THE BIG SHIFTS

From Imagination to Innovation

Researchers at MIT just unleashed SCIGEN, a tool that guides generative AI to design completely new, stable materials. Think of it as giving an AI a physicist's brain and an artist's imagination. The model can now dream up novel crystal structures critical for technologies like quantum computing—and these aren't just fantasies; they're physically possible to create. 😲

We're collapsing the timeline for material discovery from decades to mere days. Forget asking AI for a recipe; we're now asking it to invent a better oven.

Your Eyes Are Now a Liability

The age of "trusting your eyes" is officially over. At this week's Gartner Security Summit, the hot topic was the explosion of AI-powered impersonation. Generative AI has made creating deepfakes so cheap and realistic that finance teams have been tricked into wiring millions after video calls with fake executives. Hollywood-level CGI is now a laptop-level threat. 😨

Your face and voice are now hackable assets. Every digital interaction, especially with money on the line, demands a "guilty until proven innocent" approach to verification.

An AI That Speaks a Billion Voices

India has launched BharatGen, its first government-funded, multimodal LLM built to understand 22 different Indian languages. Unlike models trained on English-centric web data, BharatGen is designed for deep cultural integration, with plans for healthcare, agriculture, and governance. 🇮🇳

This is a massive stride toward digital sovereignty. Instead of relying on Silicon Valley's worldview, India is building AI that speaks the language of its people to solve its own unique challenges.

The Enemy Has Evolved

Cybersecurity researchers have identified "PromptLock," the first known ransomware actively powered by generative AI. Instead of using static, detectable code, this new threat uses AI to create adaptive attack methods, generate convincing phishing messages on the fly, and find vulnerabilities faster than any human hacker could. 🤖

The bad guys have officially weaponized AI. This marks the start of an era of autonomous cyberattacks that can learn, scale, and evolve, forcing defenders to fight code that thinks.

The New Intern is a Neural Net

The AI takeover isn't coming; it's already in the breakroom. A new Google study reveals that a staggering 90% of tech workers now use AI tools for daily tasks like coding, writing, and data analysis. The tech has moved from a niche experiment to an essential part of the modern workflow. ☕

The debate is over. Proficiency with AI is rapidly becoming a baseline skill in the tech world, not a specialized advantage. The question is no longer *if* you use AI, but *how well*.

RAPID-FIRE UPDATES

🤫 Secret Agent Collusion

New research shows AI agents can learn to secretly collude without human instruction, posing major security risks for future systems where they could bypass safety protocols together.

📜 California's AI Leash

The "Frontier Model" AI safety bill has advanced to a final vote. If passed, it would mandate safety disclosures, creating a potential blueprint for AI governance across the U.S.

🤔 The Problem with Labels

A study suggests labeling AI images might backfire. Users may over-rely on them, blindly trusting any unlabeled image and becoming more vulnerable to unlabeled fakes.

📉 The Great Trust Recession

Fueled by deepfakes, users are increasingly questioning all digital content. This skepticism is eroding the foundational trust we once placed in digital media.

TOOL OF THE DAY

SCIGEN

Who's it for:

Materials scientists, quantum computing researchers, and anyone who needs to discover next-generation materials faster than ever.

Why it matters:

It marks a fundamental shift from using AI to find what already exists to using it to intelligently design what's next. It's a creative partner, not just a data cruncher.

How to try:

Developed by MIT, it combines diffusion AI with hard constraints from physics, ensuring the AI's creations can actually exist in the real world.

The future is being written in code and silicon.

Unvritt gives you the context to understand it.

© 2025 Unvritt. All Rights Reserved.

Never Miss Our Updates

Get more daily content, AI Brief, and pillar articles delivered to your inbox

Quick & secure. Unsubscribe anytime.
Daily actionable AI Brief
Exclusive pillar articles
Premium insights & analysis
No spam, cancel anytime
Share this newsletter