DAILY INTELLIGENCE BRIEF

The Vault - AI Edition

Thursday, June 18, 2026
15 stories
14,184 words
62 min read

Business

5
01

Midjourney Unveils Its First Hardware Product: A 60-Second Full-Body Ultrasound Scanner

The company best known for conjuring photorealistic images from text prompts now wants to image something far more personal: the inside of your body.

4 minRead →
02

JPMorgan Cuts Off Hong Kong Staff From Anthropic's Claude, Following Goldman Sachs

JPMorgan pulled Anthropic's Claude from its Hong Kong staff's approved-AI list over licensing terms, following Goldman Sachs.

4 minRead →
03

NeuralTrust Raises $20 Million to Secure the Enterprise's Growing Swarm of AI Agents

Most large companies cannot tell you how many AI agents are running inside their walls right now -- the blind spot NeuralTrust just raised $20 million to fix.

4 minRead →
04

ByteDance Has Quietly Become Microsoft's Biggest AI Customer, on Track to Spend Over $1 Billion a Year on Azure

Washington spent two years curbing Chinese access to US AI -- while one of the largest such relationships sat inside Microsoft's own books: ByteDance.

5 minRead →
05

South Korean and Taiwanese Suppliers Reap the AI Hardware Boom as US Curbs Sideline China

When South Korea's Kospi punched through 9,000 for the first time, the catalyst wasn't software. It was a memory chip.

4 minRead →

Llms Genai

4
06

Zhipu's Open-Weights GLM-5.2 Tops the Intelligence Index and Undercuts GPT-5.5 on Coding

A Chinese lab just shipped a model anyone can download for free that beats OpenAI's flagship on long-horizon coding tasks at roughly one-sixth the cost.

4 minRead →
07

OpenAI Upgrades GPT-Rosalind, Its Life-Sciences Model Built for Drug Discovery

OpenAI has overhauled GPT-Rosalind, the model it built for drug discovery, fusing GPT-5.5's agentic coding with sharper reasoning in medicinal chemistry, genomics, and the wet lab.

4 minRead →
08

Google's Antigravity CLI Replaces the Gemini CLI as the Legacy Model Catalog Gets Cleared

On June 18, 2026, the gemini command thousands of developers wired into their terminals and CI/CD pipelines stopped serving requests.

4 minRead →
09

'Continuous Audio Thinking' Aims to Give Large Audio Language Models a Real Inner Voice

Ask a voice assistant to name a song's genre and something lossy happens: by the time it reasons, it's thinking about a text description of sound, not the sound itself.

4 minRead →

Research

4
10

OpenAI's LifeSciBench Grades AI on Real Life-Science Research -- and the Best Model Passes Just 36%

When OpenAI handed its models 750 research problems written by working biotech and pharma scientists, the best system cleared barely a third.

4 minRead →
11

Pew: Nearly Half of US Adults Now Use AI Chatbots, but Trust in AI's Impact Stays Low

About half of US adults now report using an AI chatbot, and one in four reach for one daily -- yet they remain unconvinced the technology is good for the country.

5 minRead →
12

NAVI-Orbital Puts a Zero-Shot Vision-Language Model in Space for Autonomous Earth Observation

For the first time, an Earth observation satellite has answered a plain-English question about the world below it -- without sending a single pixel back first.

4 minRead →
13

Gaussian Mixture Attention Offers a Linear-Time Path Around the Transformer's Core Bottleneck

Every Transformer carries a hidden tax: comparing all pairs of tokens, a cost that grows with the square of the sequence length. A new preprint proposes a way around it.

4 minRead →

Policy

2
14

A South Korean Telecom's Alleged China Ties Helped Trigger Washington's Crackdown on Anthropic's Mythos

When Anthropic pulled its two most powerful models offline on June 12, it framed it as a regulatory accident. Reporting suggests it began with one customer.

4 minRead →
15

At the G7's Evian Summit, AI Chiefs and World Leaders Strike a Voluntary Deal Centered on Youth Safety

When the 52nd G7 summit closed on June 17 in Evian, the most consequential photograph wasn't of seven heads of state. It was of a lunch table.

4 minRead →