xAI drops its Grok 2.5 model weights

PLUS: Meta's Midjourney deal, Apple eyes Google's Gemini, and a major AI browser flaw

Author

Gen AI Art
August 26, 2025 • Estimated Reading Time: 9 minutes

Together with

xAI has made a major play in the open-source community, releasing the model weights and full system prompts for its Grok 2.5 model. This move makes the model's core components publicly available for developers to experiment with.

This release provides the developer community with another powerful tool and puts more pressure on closed-model labs. As xAI signals a continued commitment to this strategy, will it force a broader industry shift toward open-source releases?

Today in AI:

  • xAI opens up its Grok 2.5 model

  • Meta’s deal to license Midjourney’s tech

  • A major security flaw in AI browsers

Business as usual? No thanks.

The problem with most business news? It’s too long, too boring, and way too complicated.

Morning Brew fixes all three. In five minutes or less, you’ll catch up on the business, finance, and tech stories that actually matter—written with clarity and just enough humor to keep things interesting.

It’s quick. It’s free. And it’s how over 4 million professionals start their day. Signing up takes less than 15 seconds—and if you’d rather stick with dense, jargon-packed business news, you can always unsubscribe.

What’s new? In a significant move for open-source AI, xAI has released Grok 2.5, making its model weights and full system prompts publicly available for developers.

What matters?

  • The release provides direct access to the model's core components, including the model weights and system prompts that define its behavior.

  • Developers can now download and experiment with the model directly from the popular open-source platform Hugging Face.

  • Looking ahead, Elon Musk is already teasing an open-source version of Grok 3 in approximately six months, signaling a continued commitment to this strategy.

Why it matters?

This release empowers developers and researchers with another powerful tool, accelerating innovation across the community. The move also raises the competitive stakes for other major AI labs to follow suit with more open models.

What’s new? Meta has announced a deal to license Midjourney's powerful image and video models. The move is designed to integrate the startup's renowned "aesthetic technology" directly into Meta's suite of social media apps.

What matters?

  • The deal positions Meta to keep pace with rivals like OpenAI and Google, which have already launched their own popular text-to-video models, Sora and Veo.

  • By licensing Midjourney's leading "aesthetic technology," Meta can quickly embed top-tier visual generation capabilities into its products without building them from the ground up.

  • This integration means users on platforms like Instagram and Facebook could soon get powerful new tools for creating high-quality images and videos, directly within the apps.Why it matters?

This partnership signals a major trend of collaboration between social media giants and specialized AI labs. It will put advanced generative tools into the hands of billions, fundamentally reshaping the future of digital content creation.

Former Zillow exec targets $1.3T market

The wealthiest companies tend to target the biggest markets. For example, NVIDIA skyrocketed nearly 200% higher in the last year with the $214B AI market’s tailwind.

That’s why investors are so excited about Pacaso.

Created by a former Zillow exec, Pacaso brings co-ownership to a $1.3 trillion real estate market. And by handing keys to 2,000+ happy homeowners, they’ve made $110M+ in gross profit to date. They even reserved the Nasdaq ticker PCSO.

No wonder the same VCs behind Uber, Venmo, and eBay also invested in Pacaso. And for just $2.90/share, you can join them as an early-stage Pacaso investor today.

Paid advertisement for Pacaso’s Regulation A offering. Read the offering circular at invest.pacaso.com. Reserving a ticker symbol is not a guarantee that the company will go public. Listing on the NASDAQ is subject to approvals.

What’s new? A critical security vulnerability in AI-powered browsers allows attackers to hijack the AI to steal private data, like emails and passwords, by hiding malicious instructions on regular websites.

What matters?

  • Researchers at the browser company Brave exposed a vulnerability that tricks an AI browser into executing hidden commands when a user asks it to perform a simple task, like summarizing a page.

  • This attack works because large language models fundamentally treat all text equally, meaning they can't distinguish between a user's prompt and malicious instructions embedded in a website's content.

  • The issue highlights why major players are cautious, with OpenAI currently offering its advanced agent only through a sandboxed cloud instance to prevent this exact type of security breach.

Why it matters?

This isn't a simple bug but a foundational challenge for AI agents that can interact with the web and personal data. Until new security protocols are established, users should be cautious about granting AI browsers access to sensitive accounts.

Everything else in AI

Google published a detailed report on its AI energy use, revealing a typical Gemini text query consumes about 0.24 watt-hours, substantially lower than many public estimates.

The U.S. government converted $8.9B in CHIPS Act grants into equity, making it the third-largest shareholder in Intel to bolster domestic chip manufacturing.

YouTube confirmed it is applying AI upscaling enhancements to videos on its platform, including Shorts, without notifying or seeking consent from the original creators.

ZoomInfo revealed how it used AI agents to automate content creation, successfully reducing its product marketing team from 26 people down to just two who now manage the systems.

Essential AI Guides - Reading List:

Your opinion matters!