AI Digest · Archived

AI Digest

An archived edition. For the current digest, see the latest.

Edition of June 26, 2026 · 42 sources · 605 articles · 48h window ← Latest digest · All past digests Subscribe via RSS

  1. OpenAI delays GPT-5.6 after a White House safety request.

    Sam Altman reportedly told staff that GPT-5.6 will ship only as a limited preview to select partners, after the Trump administration asked OpenAI to slow the rollout over security concerns. Separately, OpenAI pushed a free-tier upgrade to GPT-5.5 Instant that it says better understands user intent and is already live in the API.

    Sources The Verge ·TechCrunch ·VentureBeat

  2. OpenAI and Broadcom unveil a custom inference chip, "Jalapeño."

    The two companies announced silicon purpose-built for running LLM inference at scale, OpenAI's first in-house chip effort, aimed at easing the compute crunch and cutting cost per token. It signals OpenAI wants to be seen as a full-stack player, not just a model maker dependent on Nvidia.

    Sources OpenAI ·Ars Technica ·The Register

  3. The AI-driven memory shortage is forcing sweeping price hikes.

    Apple raised Mac and iPad prices and Microsoft hiked Xbox prices, both blaming surging memory and storage costs, while Micron locked in historically high memory prices for five years. Soaring AI datacenter demand for RAM is now rippling out to consumer hardware across the industry.

    Sources Ars Technica ·The Verge ·The Register ·TechCrunch

  4. Anthropic accuses Alibaba of the largest Claude-cloning attack to date.

    Anthropic alleges Alibaba used roughly 25,000 accounts to mine Claude across 28.8 million exchanges to steal its capabilities, and says the company should be punished. The fight lands as data shows Anthropic's Claude is steadily winning over paying consumers in a market long dominated by ChatGPT.

    Sources Ars Technica ·Wired ·TechCrunch

  5. IBM claims the first sub-1-nanometer chip, extending Moore's Law.

    IBM built a prototype packing around 100 billion transistors on a fingernail-sized area, roughly double the density of its previous record, using stacked "nanostack" transistors. The company says the process could scale toward 1 Angstrom and keep performance and efficiency gains coming for another decade.

    Sources MIT Tech Review ·Ars Technica ·The Register

  6. Agents are moving from chat to actually doing the work.

    OpenAI published research arguing agents now handle longer, more complex tasks, and says its own employees are shifting from chat to agents like Codex. Notion shut down its email app entirely as users hand inboxes to AI agents, while InfoQ reports agents climbing up the software lifecycle into code review and PRD governance.

    Sources OpenAI ·The Register ·TechCrunch ·InfoQ