Today in AI — 7 July 2026

Today’s throughline is control: who can crawl, train, impersonate, execute, and sell the next layer of AI. The model race still matters, but the sharper fight is moving into settings panels, agent platforms, security workflows, and supply chains.

Permission becomes product design

Cloudflare and Google are both turning AI data access into a product decision: site owners get finer crawler controls, while users need to opt out if they do not want ordinary Google use feeding AI training. Apple and China show the consumer side of the same issue: voice, personality, and emotional design are becoming configurable, but also governable.

• Cloudflare replaces its blanket AI bot block with granular controls for search, training, and agent crawlers — The Decoder
• If you use Google, you’re training its AI. Here’s how to opt out. — TechCrunch
• You can now customize Siri’s pace and expressivity in the latest iOS 27 beta — TechCrunch
• China forces its biggest AI platforms to shut down humanlike chatbot personas — The Decoder

Agents enter the operating layer

The agent story is splitting in two: builders want agents that can act inside real workflows, while defenders now need to assume that attackers can automate more of the chain. That makes product boundaries matter; the agent is no longer a chat box, it is an execution surface.

• Vercel CEO Guillermo Rauch on the fight to split off models from agents — TechCrunch
• Government of Alberta uses Claude to find and fix cybersecurity vulnerabilities across government systems — Anthropic
• Reddit is using LLMs to solve a problem LLMs largely created — TechCrunch
• The ‘first’ AI-run ransomware attack still needed a human — TechCrunch
• JADEPUFFER is the first agentic ransomware operation and it exposes old security sins at machine speed — The Decoder

Model advantage keeps compressing

If GPT-4’s dominance lasted a year and today’s top models barely survive seven weeks at the top, product teams should stop treating model choice as a permanent architecture decision. Open models and cheaper coding agents add pressure from below, which favours systems that can swap providers without rewriting the product.

• GPT-4's dominance lasted a year while today's top models barely survive seven weeks at the top — The Decoder
• Tencent releases Hy3 open-source model that allegedly matches models up to five times its active size — The Decoder
• Zhipu AI launches ZCode to challenge Claude Code and OpenAI Codex at a fraction of the cost — The Decoder

The stack is still physical and regional

AI demand keeps spilling into memory makers, server racks, and startup hubs. The useful correction here is simple: software may move fast, but capital, supply, and geography still decide who can turn capability into revenue.

• US investors will soon get access to SK Hynix, another memory maker riding the AI boom — TechCrunch
• Nvidia's Kyber NVL144 reportedly pushed back more than a year, Asian suppliers drop — The Decoder
• Station F ramps up as a launchpad for Europe’s hottest AI startups — TechCrunch

The takeaway for builders: design for churn, permission, and containment first; the model is only one part of the product now.

---