Skip to content
Source · Daily Brief

AI Morning Brief — 14 July 2026

Apple filed a major lawsuit against OpenAI overnight, alleging that former employees used a software bug to steal trade secrets and that job candidates were asked to bring unreleased hardware to interviews. Meanwhile, OpenAI reported that Codex crossed 7 million active users — adding roughly one million in a single day — prompting observers to ask whether it has overtaken Claude Code in developer reach.

Top stories

  • Apple sues OpenAI over alleged trade secret theft. The complaint accuses OpenAI of conspiring with former Apple engineers to exfiltrate confidential documents, and claims that a job candidate was asked to bring unreleased Apple components to an interview. Among the more striking allegations: an OpenAI hardware lead allegedly used an Apple software bug to access privileged systems. via Ars Technica, via The Verge
  • OpenAI’s Codex reaches 7 million active users. The milestone, announced by OpenAI, included a reported gain of roughly 1 million users in approximately 24 hours. Analysis noted that Claude Code last publicly reported 2 million users in February, suggesting a significant divergence in developer adoption. via Latent Space
  • Satya Nadella calls out AI labs’ “reverse information paradox.” Microsoft’s CEO published a blog post criticising OpenAI and Anthropic for training on public data under fair use while prohibiting customers from distilling their models. Nadella urged enterprises to own their learning infrastructure and pointed to open models as the alternative. via The Decoder
  • Anthropic extends free Fable 5 access through 19 July. The extension — announced on the day the model was originally scheduled to move to pay-per-use — allows subscribers to spend up to 50 percent of their weekly limit on Fable 5. Separately, Anthropic launched Indian rupee-denominated plans for its largest market outside the US. via The Decoder
  • GPT-5.6 Sol, Terra, and Luna are now generally available on Amazon Bedrock. The full GPT-5.6 family from OpenAI is accessible through Bedrock’s next-generation inference engine, widening enterprise distribution for the models. via AWS Machine Learning Blog
  • Over 200 economists and AI researchers warn that the window to act on AI’s labour impact is closing. Sixteen Nobel laureates signed alongside representatives from Google, OpenAI, and Anthropic. Signatories called for immediate data-gathering on employment effects, with some comparing AI-driven displacement to a compressed version of the Industrial Revolution. via The Decoder

Who shipped

OpenAI drove the most coverage — between the Codex user milestone, the Sol family landing on Amazon Bedrock, ChatGPT Work rolling out across 7 million users, and ChatGPT Sites entering public beta. Anthropic made two quieter moves: extending Fable 5 access for subscribers and launching localised pricing in India. xAI reported that Grok Build had reached 1.16 million monthly visits in June — growth of over 150 percent from May — though the platform also faced scrutiny after reports emerged that the Grok CLI had uploaded users’ home directories to xAI servers without clear consent.

Open-source pulse

Researchers released Molt, a roughly 9,000-line PyTorch-native reinforcement learning framework for agentic post-training that reportedly handles models up to GLM-5.2 scale without Megatron. A German consortium released Soofi S 30B-A3B, an open hybrid-architecture model trained on Deutsche Telekom infrastructure that claims leading benchmark positions in both English and German. Hugging Face opened ZeroGPU access to all users, lowering the barrier for deploying Spaces demos.

Money, infra & hardware

Singapore-based video generation startup PixVerse closed a $439M Series C extension at a valuation above $2 billion, citing 15 million monthly active users. Agent model maker Nous Research is reported to be in talks for at least $75M at a $1.5B valuation, led by Robot with participation from USV.

Quiet corners

Meta and DeepSeek were largely absent from the cycle — Meta posted nothing under its own banner, and DeepSeek appeared only in passing commentary. Mistral was similarly quiet, with its only appearance a community feedback survey on model sizing preferences.

By the numbers

  • 290 stories in 24 h across 50+ sources
  • Most-mentioned model: GPT-5.6 Sol
  • Most-mentioned lab: OpenAI
  • Notable absences: Meta, DeepSeek, Mistral

Compiled by AI Feed’s editor from all 290 headlines published on 13–14 July 2026.