AI News, Jul 7, 2026: Meta launches Muse Image generator using Instagram photos for training

Today's updates highlight a shift toward practical deployment, with companies cutting costs, expanding mobile agents, and facing privacy pushback.

1. 01
   Meta launches Muse Image generator using Instagram photos for trainingWIRED AI

   Meta released its Muse Image model, which powers image generation across its apps and trains on public Instagram photos unless users manually opt out.

   Why it matters

   Unless you opt out, Meta can now use your public Instagram photos to generate new AI images, raising significant privacy concerns.
2. 02
   Discord fixes AI moderation bug that wrongfully banned usersTechCrunch

   Discord confirmed it resolved an AI moderation glitch that had been incorrectly banning users for harmless images since May, affecting hundreds of accounts.

   Why it matters

   The error highlights the risks of automated moderation, which can lock users out of their accounts for sharing completely harmless images.
3. 03
   Microsoft replaces external AI models with in-house tech to cut costsThe Decoder

   Microsoft is replacing third-party AI models with its own MAI models in products like Excel and Outlook to reduce reliance on expensive external providers.

   Why it matters

   By phasing out OpenAI and Anthropic models in Copilot, Microsoft aims to lower operational costs, which could eventually lead to cheaper consumer pricing.
4. 04
   Anthropic expands Claude Cowork AI agent to mobile and webThe Decoder

   Anthropic released its background-running AI agent, Claude Cowork, on mobile devices and web browsers after previously limiting it to its desktop application.

   Why it matters

   You can now run background tasks and receive updates on your phone even when your computer is closed, making AI agents more practical for daily work.
5. 05
   China considers export limits on its most powerful AI modelsThe Decoder

   Chinese authorities are reportedly planning to restrict foreign access to advanced AI models from companies like Alibaba and ByteDance.

   Why it matters

   Restrictions could limit global access to competitive, low-cost Chinese AI models, forcing businesses to rely on more expensive Western alternatives.
6. 06
   Anthropic releases tool to read Claude's internal reasoningThe Decoder

   Anthropic introduced Jacobian Lens, an analysis tool that can read Claude's newly discovered internal working memory to see how it evaluates prompts.

   Why it matters

   Opening the "black box" of AI memory helps developers and researchers understand how models think, making AI systems more transparent and predictable.
7. 07
   Chinese AI startup Deepseek is designing its own chipsThe Decoder

   Chinese AI startup Deepseek is reportedly developing its own proprietary semiconductor chips to power its AI models.

   Why it matters

   Custom hardware could help Chinese firms bypass US export restrictions and lower the cost of training advanced AI models.