- The week’s most-surfaced arXiv papers cluster on making AI agents reliable rather than simply larger.
- “Escaping the Self-Confirmation Trap” proposes an execute–distill–verify paradigm to stop agents from misremembering wrong-but-self-consistent trajectories as successful experience, while “OpenThoughts-Agent” offers an open data-curation pipeline for training broadly capable agents across diverse tasks.
Snapshot — June 25, 2026
35 stories
- Amazon said it will invest a further $13 billion through 2030 to expand AWS data-center capacity in Mumbai and Hyderabad, announced after CEO Andy Jassy met India’s Prime Minister Modi.
- The commitment brings Amazon’s cumulative India pledges to roughly $48 billion, tracking a broader race among hyperscalers to secure AI compute footprint in the country.
- Amazon, Microsoft and other tech firms joined RAISE US, a new bipartisan workforce nonprofit led by former Commerce Secretary Gina Raimondo (CEO) and former Indiana Gov.
- Eric Holcomb (co-chair).
- The group aims to partner with governors and employers to retrain workers displaced by AI, targeting $1B in multi-year commitments — more than half already secured.
- Anthropic has accused Alibaba of carrying out the largest known model-distillation attack against Claude, alleging the creation of roughly 25,000 fake accounts to generate nearly 29 million conversations used to train competing models.
- Anthropic disclosed the claim in a letter to Senators Tim Scott and Elizabeth Warren of the Senate Banking Committee, first reported by CNBC.
- Apple said it will raise prices on certain MacBooks and iPads by up to $300, and Microsoft announced Xbox console price increases effective August 1, with both citing surging memory and storage chip costs.
- “We have never seen a component price increase this much, this quickly,” Apple said.
- The increases trace directly to AI data-center expansion competing for memory and storage capacity — a sign that AI infrastructure costs are now reaching consumers.
Berkeley RDI *(No new Berkeley RDI emails found for 2026-06-25)*
Business Insider *(No new Business Insider emails found for 2026-06-25)*
CIO Dive - [2026-06-25] [EXTERNAL] The AI visibility gap hiding in plain sight. - [2026-06-25] [EXTERNAL] June 25 - AI coding costs set to climb | Linux Foundation preps agentic standard
*Coverage from newsletter subscriptions for 2026-06-25*
DealBook (Andrew Ross Sorkin / NYT) - [2026-06-25] [EXTERNAL] DealBook: Hot New York summer
- Despite cooling public-market sentiment, private capital kept flowing into applied AI.
- Norm AI, which embeds legal and compliance rules into AI agents for regulated enterprises, disclosed a $120M raise across nineteen investors.
- Taktile, an AI decision platform for financial institutions, closed a $110M Series C led by Goldman Sachs Alternatives' growth-equity arm, with Balderton, Index, Tiger Global and Y Combinator participating — signaling continued appetite for vertical, workflow-embedded agents.
- York lab General Intuition closed a $320 million Series A at a $2.3 billion valuation to scale models trained on millions of hours of human gameplay clips, betting that action data yields agents with more human-like “intuition.” The company is applying the same underlying model to both in-game agents and physical robots.
- IBM said its Albany research lab produced the first chip technology to operate below 1 nanometer — a 0.7nm (7-angstrom) transistor built in three dimensions via a "nanostack" architecture, packing ~100 billion transistors onto a fingernail-sized area.
- Though IBM exited chip manufacturing years ago, the result positions its research ahead of TSMC, Samsung and Intel on transistor scaling and points to a longer runway for compute density as AI demand surges.
- Domyn (formerly iGenius) CEO Uljan Sharka said the company will release a fully open-source "frontier" model within a year, developed through its EUROPA consortium with Germany’s Fraunhofer-Gesellschaft under the European Commission’s Frontier AI Grand Challenge.
- The effort positions Domyn alongside Mistral and OVHcloud as Europe seeks sovereign alternatives — context sharpened by Italy and Czechia restricting remote use of DeepSeek and by U.S. export controls on Anthropic’s models.
- Per Financial Times reporting summarized this Thursday, Meta has shifted roughly half of content-review tasks to large language models and aims to exceed 90% AI-driven moderation for certain categories before year-end — an acceleration of a previously multi-year, human-in-the-loop plan.
- The shift accompanies broader cost cuts (≈8,000 roles eliminated) and Zuckerberg’s "personal superintelligence" push.
- Microsoft fell ~3.5% after Stifel lowered its price target to $400 from $415 and held a Hold rating, citing concern that margin expectations may be unrealistic as AI infrastructure investment climbs.
- Azure growth remains the key support, but the debate has shifted to whether cloud and AI demand can offset the rising cost of building and operating capacity.
- Researchers from MIT and Microsoft developed a system that lets developers describe agentic workflows in plain language, then automatically optimizes how those workflows are implemented — addressing the fragmentation that forces cloud operators to over-provision compute.
- Lead author Gohar Chaudhry (MIT EECS) frames it as a cloud-infrastructure-layer fix rather than a model-layer one, targeting the orchestration waste that emerges when multiple models and tools are chained.
No items met the strict 24-hour publication bar. Notable June policy stories — including state-level lawsuits and government model-access disputes — are all dated earlier than June 24 and were deliberately excluded to keep this digest current.
- OpenAI made Codex Remote generally available on every ChatGPT plan, letting users start or continue work on a connected Mac or Windows host and approve actions from the mobile app, with authenticated one-to-one QR device pairing.
- It also added a DigitalOcean Droplet Workspace plugin so Codex can provision and connect to a remote sandbox.
- OpenAI updated GPT-5.5 Instant — the default model for free ChatGPT users — with stronger intent inference, better handling of complex multi-constraint requests, and improved shopping and local-recommendation results.
- The company framed it as a move away from models that need heavy prompting toward systems that preserve user goals across turns.
- A new OpenAI paper, “The Shift to Agentic AI: Evidence from Codex” (co-authored with Columbia, Duke, and the University of Pennsylvania), reports that 97.9% of OpenAI employees now use Codex — up from ~40% in August 2025 — with non-technical departments like Legal and Recruiting adopting it as their primary tool.
- OpenAI revealed its first custom accelerator, "Jalapeño," co-designed with Broadcom (engaged last October) and architected to reduce data movement while balancing compute, memory and networking for inference.
- Broadcom’s Hock Tan and Charlie Kawwas presented the first unit to Sam Altman and Greg Brockman this week.
- Patronus AI, founded by former Meta researchers Anand Kannappan and Rebecca Qian, raised a $50M Series B led by Greenfield Partners (with Lightspeed, Notable Capital, Datadog and Samsung Ventures), bringing total funding to $70M.
- The startup builds simulated "world model" environments to evaluate and harden autonomous agents before deployment in high-stakes domains such as trading, healthcare and drone automation.
PitchBook - [2026-06-25] [EXTERNAL] SEC wants PE to show its work
Scope: Only items confirmed published June 24–25, 2026 (last 24 hours) are included; undated and older stories were excluded.
- Bloomberg reported SK Hynix is seeking to raise roughly $29.4B in a US listing, with trading expected July 10 and proceeds earmarked for additional high-bandwidth memory (HBM) capacity — the critical bottleneck for AI accelerators.
- As the leading HBM supplier to Nvidia’s H100/H200/GB200 families, SK Hynix’s listing is a barometer of memory-sector confidence in the sustained AI infrastructure build-out.
- Sources scanned: Companies — Nvidia, Google/DeepMind, OpenAI, Anthropic, Mistral, Cursor, Replit, Meta, Apple, Amazon, Cerebras, Microsoft, Palantir, Oracle, IBM, Tencent, Baidu, Databricks, xAI, Alibaba, Huawei, SenseTime, DeepSeek.
- Universities — UC Berkeley, Stanford, MIT, Purdue, Georgia Tech, Princeton, Carnegie Mellon, University of Washington, Cornell, UT Austin, UC San Diego.
The Information *(No new The Information emails found for 2026-06-25)*
The Tactical Allocation Letter *(No new The Tactical Allocation Letter emails found for 2026-06-25)*
This file catalogs email subjects received. Full article extraction requires individual email processing via merge_publications.py.*
Wall Street Journal / WSJ *(No new Wall Street Journal / WSJ emails found for 2026-06-25)*
Per a staff memo reported by The Information, OpenAI will initially release GPT-5.6 only to parties approved by the federal government, with leaders “approving access customer by customer during this preview period” before a wider release weeks later. Altman reportedly told employees this “is not our preferred long term model” and that OpenAI will push for a more sustainable approach — signaling an emerging de facto licensing posture for frontier releases.
- OpenAI reportedly plans to share its newest model, GPT-5.6, only with a select group of partners during a preview — because the Trump administration asked it to, per The Information.
- CEO Sam Altman reportedly told staff the government would approve access "customer by customer," with a broader release hoped for a couple of weeks later if the limited rollout goes well.
WSJ Pro CyberSecurity *(No new WSJ Pro CyberSecurity emails found for 2026-06-25)*
WSJ Wealth Advisor *(No new WSJ Wealth Advisor emails found for 2026-06-25)*