How to Run Gemma 4 locally with OpenClaw 🦞

Welcome to another edition of Horizon AI,

In today's issue, we show you how to run Google's new Gemma 4 model locally and connect it with OpenClaw in just a few minutes.

Let’s jump into it!

Read Time: 4.5 min

Here's what's new today in the Horizon AI

• Anthropic Lets Big Tech Test Powerful Mythos AI Model

• OpenAI, Anthropic, and Google Team Up Against AI Copying

• AI Tutorial: Run Gemma 4 locally with OpenClaw

• AI Tools to check out

• AI Findings/Resources

• The Latest in AI and Tech 💡

AI News

ANTHROPIC

Anthropic Lets Big Tech Test Powerful Mythos AI Model

Anthropic is giving select companies early access to its unreleased Mythos model to help prepare for potential cybersecurity risks.

Details:

• Through an initiative called Project Glasswing, companies like Nvidia, Apple, Amazon, Microsoft, and Cisco can test Mythos to hunt for flaws in their products and share findings with industry peers.

• Anthropic will use feedback from these partners to build safety guardrails before any wider release.

• The company claims the model has found thousands of exploitable vulnerabilities, including decades-old bugs and flaws missed by traditional automated testing tools, despite being a general-purpose AI model not specifically developed for cybersecurity.

• Anthropic says the model's "strong agentic coding and reasoning skills" are behind its cybersecurity advances.

Anthropic also said it has been in "ongoing discussions with US government officials about Claude Mythos Preview and its offensive and defensive cyber capabilities," despite its recent clash with the Trump administration.

Source

TOGETHER WITH MINTLIFY

AI agents now read your docs almost as much as humans do.

Mintlify analyzed 790 million requests across its documentation platform. The finding: AI coding agents account for 45.3% of all traffic, nearly tied with traditional browsers at 45.8%.

Two tools are driving almost all of it:

• Claude Code: 25.2% of total traffic, more requests than Chrome on Windows

• Cursor: 18% of total traffic

• Together they account for 95.6% of all identified AI agent traffic

The rest of the field, OpenCode, Trae, ChatGPT, and NotebookLM, is showing up but nowhere close.

One caveat: OpenAI's Codex doesn't send an identifiable user-agent header, so the real agent percentage is likely even higher.

The takeaway for anyone maintaining developer docs: your documentation now serves two audiences. Structure and machine-readability matter as much as clarity for human readers.

Try Mintlify today!

AI RACE

OpenAI, Anthropic, and Google Team Up Against AI Copying

Top AI labs are joining forces to stop Chinese competitors from copying their models using advanced techniques.

Details:

• OpenAI, Anthropic, and Google are sharing data through the Frontier Model Forum to detect “adversarial distillation,” where outputs from one model train a cheaper copy.

• The practice has become a major issue, with U.S. authorities estimating it costs AI companies billions in lost revenue each year.

• Companies have identified groups like DeepSeek, Moonshot, and MiniMax as using increasingly sophisticated methods to extract model behavior.

This marks a rare moment of cooperation among rivals who have had plenty of friction between them lately, as the industry recognizes that protecting AI systems is becoming just as important as building them.

Source

AI Tutorial

Run Gemma 4 locally with OpenClaw

1. Install Ollama from https://ollama.com/download

2. Launch OpenClaw with Gemma 4 as the backend via Ollama.

This will automatically install OpenClaw on your system and launch it with Gemma 4 26B A4B.

AI Tools to check out

🔥 Naive: Hire AI employees to create and run your company.

🖱️ Clicky: An AI teacher that lives as a buddy next to your cursor.

⚡ Dex: Ask your data anything. Get answers in seconds.

🚀 Sokosumi: A marketing productivity tool that lets teams work with specialized AI marketing agents - so real work gets done faster.

🔊 Willow: AI-powered voice dictation that's so powerful it can replace your keyboard.

TOGETHER WITH MINTLIFY

Are you tracking agent views on your docs?

AI agents already outnumber human visitors to your docs — now you can track them.

See your AI traffic!

AI Findings/Resources

🤯 An unknown model "HappyHorse" has taken the No. 1 spot on Artificial Analysis' text-to-video leaderboard, surpassing even ByteDance's Seedance 2.0.

💀 Someone made a digital whip to make Claude work faster

🤔 Sam Altman's coworkers say he can barely code and misunderstands basic machine learning concepts

🦠 Scientists invented a fake disease. AI told people it was real

🛍️ AI is changing how small online sellers decide what to make

The latest in AI and Tech

Musk updates OpenAI lawsuit so the nonprofit receives any trial winnings

Elon Musk has updated his lawsuit against OpenAI and Microsoft, asking that any potential damages (reportedly over $150 billion) be awarded to OpenAI's nonprofit foundation instead of himself. Musk is also seeking the removal of Sam Altman from the foundation's board.

Study finds Google AI Overviews are accurate most of the time, but harder to verify

AI startup Oumi analyzed over 4,000 queries and found accuracy improved from 85% with Gemini 2 to 91% with Gemini 3. However, even small error rates could translate into millions of incorrect answers given Google’s massive search volume.

The study also flagged a growing issue with verifiability. More than half of correct answers from Gemini 3 weren’t fully supported by the sources Google cited, making them harder to fact-check.

Microsoft open-sources Harrier embedding model

Microsoft’s Bing team has released “Harrier,” a new open-source embedding model designed to improve how AI systems search, retrieve, and organize information. The model supports over 100 languages, features a large context window, and was trained on billions of examples, including synthetic data from advanced models.

Intel joins Musk’s Terafab project

Intel is partnering with Elon Musk to help design and build the Terafab semiconductor facility in Texas, aimed at producing AI chips for SpaceX and Tesla.

Microsoft cloud exec joins Anthropic

Anthropic has hired Eric Boyd, a senior Microsoft executive, to lead its infrastructure efforts. Boyd previously ran Microsoft's AI Platform, which powers the deployment of large language models for both customers and internal teams.

That’s a wrap!

Thanks for sticking with us to the end! Let’s stay connected on LinkedIn and Twitter.

Not subscribed yet? Sign up here and send it to a colleague or friend!

See you in our next edition!

Gina 👩🏻‍💻