Monday, July 13, 2026

Apple sued OpenAI over allegedly stolen hardware trade secrets (400+ ex-Apple staff now at OpenAI), touching off a weekend of Musk-vs-Altman sniping. China is weighing export curbs on its own top models — the cheap Chinese-model arbitrage your team may be leaning on could tighten. OpenAI's deployment arm bought Northslope to rent out forward-deployed engineers. And the only hard deadline on the board this week is a boring but real one: DeepSeek retires its legacy API model names on Jul 24.

TLDR

Apple sues OpenAI for trade-secret theft — targets io Products and two ex-Apple designers; complaint alleges a coordinated hardware-poaching scheme (Jul 10) Trending
DeepSeek deprecates deepseek-chat / deepseek-reasoner API names on Jul 24, 15:59 UTC — no grace period after (deadline Jul 24) Pricing
China weighs restricting overseas access to Alibaba/ByteDance/Z.ai's most advanced models, including open-weightA model whose trained weights are published for anyone to download and run.NTIA ↗ releases (Reuters, Jul 7) Policy
Claude Code ships a built-in sandboxed browser and turns auto mode on by default for Bedrock/Vertex/Foundry (Jul 11) Agents
OpenAI's Deployment Company acquires Northslope, adding hundreds of Palantir-style forward-deployed engineers (Jul 8) Agents

Action Items

Things your engineering team should look at or do this week.

Migrate any DeepSeek calls off deepseek-chat/deepseek-reasoner before Jul 24 15:59 UTC. After that, those model names return errors with no fallback. Swap to deepseek-v4-flash (thinking on/off) or deepseek-v4-pro. (DeepSeek API docs)
Re-check your cheap-Chinese-model dependency for supply risk. With Beijing openly discussing export curbs (even on open weights) and U.S.–China legal friction rising, any production path routed to DeepSeek/GLM/Qwen should have a benchmarked fallback to a Western model before a policy change forces one. (Fortune/Reuters)
Note the Claude Code auto-mode default change if you run it in CI. As of Jul 11, auto mode is on by default on Bedrock, Vertex AI, and Foundry without the prior opt-in env var — confirm that matches your intended autonomy/permission posture. (Claude Code changelog)

Models & Tools

Claude Code adds a built-in sandboxed browser and flips auto mode on by default (Jul 11) Agents — The desktop app now bundles a sandboxed browser so Claude can pull up docs, designs, and live sites and interact with them the way it does with local dev servers. The same release makes auto mode the default on Bedrock, Vertex AI, and Foundry (no CLAUDE_CODE_ENABLE_AUTO_MODE opt-in), updates Bedrock to Opus 4.8, and fixes terminal freezes when streaming very long lists/tables/code. (Claude Code changelog)
OpenAI's GPT-Live full-duplex voice models roll out in ChatGPT (Jul 8) Trending — GPT-Live-1 and GPT-Live-1 mini replace Advanced Voice Mode with a full-duplex architecture that listens and speaks simultaneously (backchannels like "mhmm," goes quiet when you pause) and delegates hard questions to a frontier model behind the scenes. GPT-Live-1 becomes default for Go/Plus/Pro, mini for Free; it runs GPT-5.5 in the background at launch. (OpenAI)
Watch: Gemini 3.5 Pro reportedly targeting Jul 17 GA — still unconfirmed (reports Jul 13) — Multiple outlets report a Jul 17 general-availability target with a 2M-token context and Deep Think mode after Google scrapped and rebuilt the base model. Treat as not released: there is no official Google date, no model card, and no gemini-3.5-pro API listing as of today. Plan around it, don't build on it yet. (TechTimes)

APIs & Pricing

DeepSeek retires legacy API model names on Jul 24 (deadline Jul 24, 15:59 UTC) Pricing — The deepseek-chat and deepseek-reasoner aliases (kept alive as a 90-day migration bridge after V4 launched Apr 24) stop resolving next Thursday. Calls using them will error with no fallback; migrate to deepseek-v4-flash or deepseek-v4-pro. (DeepSeek API docs)

Industry & Business

Apple sues OpenAI over alleged trade-secret theft (Jul 10) Trending — Apple's complaint names OpenAI, io Products, and two former Apple employees (an ex-system electrical engineer and a former iPhone/Watch product-design VP), alleging a coordinated scheme to lift hardware designs, manufacturing processes, and supply-chain strategy for OpenAI's consumer-device push — including keeping an Apple laptop to exfiltrate confidential docs and coaching departing staff to evade security. Apple says 400+ of its former employees now work at OpenAI; OpenAI says it has "no interest in other companies' trade secrets." The suit reignited a public Musk–Altman spat over the weekend. (TechCrunch | CNBC)
OpenAI's Deployment Company acquires Northslope (Jul 8) Agents — The majority-OpenAI-owned deployment arm (launched May with ~$4B for M&A) made its second buy after Tomoro, adding hundreds of Palantir-rooted "forward-deployed engineers" who embed with customers to get AI systems running reliably in production. Northslope grew revenue 7x in 2025 pre-deal; terms undisclosed. The signal: OpenAI is selling AI adoption as a service, not just models. (Axios)

Policy & Safety

China weighs curbing overseas access to its most advanced AI models (Reuters, Jul 7) Policy — The Ministry of Commerce convened Alibaba, ByteDance, and Z.ai to discuss classifying top-tier models (proprietary and open-weight) for possible export limits — up to a bar on public release or domestic-use-only status — plus treating unauthorized disclosure of proprietary AI as a national-security violation. Nothing is decided and there's no timeline, but any curbs would raise costs for the many U.S. teams now relying on cheap Chinese models. (Fortune/Reuters | Time)
UN holds Global Dialogue on AI Governance in Geneva (Jul 6–7) Policy — The UN's inaugural governance dialogue framed frontier AI as concentrated in just two countries (US and China) and warned of "catastrophic harm" absent coordinated oversight — an early marker for where multilateral AI rules head next, even if non-binding for now. (UN News)

A quiet Monday after a blockbuster launch week — and the story has shifted from model specs to who controls the models, the talent, and the supply chain.