🗞️ AI Daily Briefing — 2026-06-26

🔥 Top Story

Google DeepMind is bleeding top talent: Noam Shazeer to OpenAI, John Jumper to Anthropic. In the span of a single week, Google lost two of its most consequential AI researchers. Shazeer — co-author of “Attention Is All You Need,” the paper that started the Transformer era — left to become Lead for Architecture Research at OpenAI. Days later, Nobel laureate John Jumper (AlphaFold) announced he’s joining Anthropic after nearly 9 years at DeepMind. Sam Altman called the Shazeer hire “only 10 years in the making.” This is the most significant talent exodus from a single lab in recent memory, and it raises real questions about DeepMind’s ability to hold the frontier. (CNBC) (TechCrunch) (Fortune)

🚀 Model & Research News

• Anthropic accuses Alibaba of massive distillation attack on Claude. In a letter to the US Senate Banking Committee, Anthropic claims operators linked to Alibaba’s Qwen lab used ~25,000 fake accounts to run 28.8 million exchanges with Claude between April and June 2026, targeting software engineering and agentic reasoning capabilities. Alibaba has not commented. (CNBC) (Tom’s Hardware)
• Fable 5 still dark — Anthropic confirms zero traffic on day 14 of US export ban. Anthropic staff publicly debunked viral claims that Fable 5 was accessible again, calling them “categorically false.” The Commerce Secretary’s congressional response deadline on the suspension is today (June 26). Claude Opus 4.8 remains unaffected and sits at #1 on LMArena. (Anthropic) (explainx.ai)
• GPT-5.6 launch window slips — now likely July. Prediction markets for a June release collapsed from ~83% to ~18%. OpenAI Chief Scientist Jakub Pachocki internally described it as a “meaningful improvement” over GPT-5.5, with rumored 1.5M token context. Still no official confirmation from OpenAI. (TechTimes) (Geeky Gadgets)
• Grok 4.3 lands on Amazon Bedrock as cheapest US-lab frontier reasoning model. xAI continues rapid iteration — Grok 4.1 Thinking still holds #1 on LMArena Text at 1483 Elo, while the newer 4.3 prioritizes lower pricing and agentic performance. (Artificial Analysis) (xAI)
• Notable research: AgileThinker and AgentHallu. AgileThinker pairs a cheap reactive head with a deeper planner for real-time reasoning in live environments (trading, robot control). AgentHallu introduces a 693-trajectory benchmark across 7 agent frameworks to test whether models can identify which step in a multi-step run caused a hallucination. (buildthisnow.com)

🛠️ Tools & Developer Updates

• Anthropic launches Claude Tag — persistent AI teammate inside Slack. Claude Tag replaces the old Slack app with an always-on agent that can be @mentioned in channels, break tasks into stages, work independently, and even proactively jump into conversations via ambient mode. Available now for Enterprise and Team customers with a 30-day opt-in window. (Anthropic) (Fortune) (VentureBeat)
• Qualcomm acquires Modular for ~$3.9B to challenge Nvidia’s CUDA dominance. Announced at Qualcomm Investor Day (June 24), the deal brings Modular’s unified AI platform — which runs models across CPU, GPU, NPU, and custom ASICs without rewrites — under Qualcomm’s umbrella. Closing expected H2 2026. (CNBC) (TechStartups)

💰 Funding & Business

• Runlayer raises $30M Series A led by Felicis, with Khosla Ventures. Andy Berman’s (ex-Zapier AI Director) enterprise AI governance platform had its round preempted by Felicis, bringing total raised to $42M. Customers include Instacart, Gusto, dbt Labs, AngelList, and Opendoor. (Fortune) (Yahoo Finance)
• Qualcomm still circling Tenstorrent for $8–10B. Jim Keller’s RISC-V AI chip startup would give Qualcomm a serious play against Nvidia. Combined with the Modular acquisition, Qualcomm is spending ~$14B this week to build an end-to-end AI stack. Talks ongoing, no final deal. (Reuters via Yahoo Finance) (TechTimes)
• OpenAI IPO confidential S-1 filed — targeting fall 2026. Goldman Sachs and Morgan Stanley are handling. Polymarket gives 76% odds for a 2026 listing. Altman told staff it could happen “within the next year” but flagged RSI progress as a variable that could delay it. Meanwhile, the House Oversight Committee is investigating Altman’s $2B+ in personal investments in companies doing business with OpenAI. (Yahoo Finance) (CryptoBriefing)
• Meta morale at “Cambridge Analytica levels” after AI restructuring. CTO Andrew Bosworth told employees the company did “an atrocious job explaining the vision” behind May’s 8,000 layoffs and 7,000 AI reassignments. Zuckerberg says no more layoffs in 2026; a company-wide hackathon is planned for July. (NPR) (HR Executive)

🐦 Notable from the Timeline

• Sam Altman on hiring Shazeer: “Only 10 years in the making… one of the people I have most wanted to work with since the very beginning of OpenAI.”
• John Jumper on leaving DeepMind: “Demis took a real chance letting me lead the AlphaFold team just six months after finishing [my PhD].” Taking time to recharge before starting at Anthropic. (X)
• Anthropic (Amol Avasare, Head of Growth) on Fable 5 access rumors: “Categorically false.” Staff confirmed exactly zero Fable/Mythos traffic on June 25.
• HN sentiment shift: Top threads are now about AI security, governance, and agent trust — not flashy demos. The market is disciplining AI, not falling out of love with it. (Hacker News)

📊 Benchmark Watch

The LMArena leaderboard (7M+ votes, 368 models) as of June 25:

• Overall #1: Claude Opus 4.8 (~1510 Elo) — top 5 within ~55 Elo points, the tightest spread on record
• Text Arena #1: Grok 4.1 Thinking (1483 Elo) — 31-point lead over nearest non-xAI model
• SWE-Bench leader: Claude Opus 4.8 (88.6%), GPT-5.5 at 82.6%
• Coding Arena: GPT-5.5 strongest in head-to-head coding play
• Frontier reasoning (GPQA): Claude Mythos Preview leads

The top tier is remarkably competitive — five models from four labs all within striking distance. (LMArena) (llm-stats.com)

🎙️ Podcast Highlights

• TBPN (now OpenAI-owned since the April acquisition) continues daily 3-hour shows. The OpenAI purchase was reportedly in the low hundreds of millions; the show generated ~$5M in ad revenue in 2025 and is on track for $30M+ in 2026. Reports to OpenAI’s Chris Lehane with claimed editorial independence. (TechCrunch)

🔗 Worth Reading

• “As top talent leaves Google DeepMind, some question if the lab can remain at the forefront” — Fortune deep-dive on the Shazeer + Jumper departures and what it means for Google’s AI position. (Fortune)
• Anthropic’s full letter to the Senate on the Alibaba distillation campaign — 25,000 fake accounts, 28.8M exchanges, and the geopolitical implications for AI export controls. (Bloomberg)
• “Why AI Agent Governance is the New SaaS” — Analysis of Runlayer’s $30M raise and what it signals about where enterprise AI spend is heading. (CTOL Digital)