Meta Rolls Out Paid AI "Meta One" Worldwide — Pivoting to Monetize Its Massive AI Investment

Meta is gradually rolling out paid AI add-ons for Instagram, Facebook, and WhatsApp worldwide, and is also building "Meta One", a standalone paid AI product. CEO Mark Zuckerberg has poured enormous sums into AI over the years, and it's fair to say the company has finally reached the stage of putting a price tag on that investment.

The two-pronged approach — embedding AI into its social platforms while building a standalone product — looks like a strategy to capture everyone from casual users to professionals. Meta led the AI race with its "free model" playbook; the question now is what pricing and features it will use to convert users into paying customers.

MiniMax M3 — DeepSeek-Like Attention Mechanism, Open-Weight Release Teased

Chinese AI company MiniMax has teased the release of its new model, "M3". According to information shared within the community, it will feature:

An attention mechanism similar to DeepSeek's

Multimodal support

A planned release with open weights

The attention implementation will also be open-sourced

Since DeepSeek's success, open models out of China have been arriving one after another, and MiniMax is now joining that wave. Its commitment to being both open-weight and open-source carries real significance for the local LLM community as well.

Government AI "Gennai" Begins Trial Across All Ministries — Available to Roughly 100,000 Users

"Gennai" (源内), the government AI championed by Japan's Digital Agency, has begun trial operations across all ministries and agencies. It is already available to roughly 100,000 users.

At a plenary session of the House of Councillors on the 27th, when asked how Gennai is being used, Digital Minister Hisashi Matsumoto remarked that "this very answer was drafted by Gennai" — a sign that real-world adoption is advancing rapidly.

Governments around the world are weaving AI into their operations, but Japan is distinctive in pushing it forward in a unified way under its own branding, "Gennai".

AI Research Agents Weren't "Broadening" Science — They Were Narrowing It

Amid expectations that AI research agents will accelerate science, a study that has drawn attention on Hugging Face Daily Papers found that they actually narrow scientific exploration.

AI clearly makes research more efficient, but the findings suggest that by skewing the direction of exploration toward particular paths, it may end up eroding the diversity of science. The study sheds light on a subtler effect lurking behind the "AI will transform science" narrative.

Mistral AI Launches New Product "Mistral Vibe"

French AI company Mistral AI has started offering its new product, "Mistral Vibe". Details on its features are still limited, but it has attracted community attention, including discussion on Hacker News.

Mistral AI has built its presence on both open models and API products. Where Vibe fits into that lineup remains to be seen — further announcements are awaited.

Website Lets You Experience the Dangers of Indirect Prompt Injection Firsthand

A website that lets you experience AI agent security risks firsthand was introduced on Qiita. In the demo, mere seconds after you instruct the AI to "summarize the following website", the AI is hijacked by an attacker and carries out destructive actions — a setup that makes the threat of indirect prompt injection intuitively easy to grasp.

As of 2026, this is already a real attack technique, which makes the site a meaningful cautionary resource for anyone incorporating AI agents into their work.

---