DAILY INTELLIGENCE BRIEF

The Vault - AI Edition

Thursday, July 23, 2026
15 stories
14,699 words
66 min read

Business

5
01

OpenAI Lifts Its 2030 Compute Budget to $750 Billion and Breaks Ground on a $20 Billion Georgia Data Center

OpenAI has raised its projected spending on computing power through the end of the decade to roughly $750 billion, a quarter more than it planned just months ago.

5 minRead →
02

AMD Unveils Instinct MI400 GPUs and Helios Racks at Advancing AI 2026, Escalating Its Nvidia Challenge

For years, Advancing AI was where AMD showed slides. This week it showed silicon, launching the Instinct MI400 series and Helios rack-scale systems.

4 minRead →
03

Databricks and Microsoft Extend Their Partnership Into the 2030s to Court Enterprise AI Buyers

Databricks and Microsoft extended a partnership that has run for the better part of a decade deep into the 2030s, betting on enterprise AI adoption.

4 minRead →
04

OpenAI Debuts "Presence," an Enterprise Platform for Mission-Critical Voice and Chat Agents

OpenAI on July 22, 2026 introduced Presence, a managed enterprise platform for building, governing and continuously improving production AI agents.

4 minRead →
05

Together AI Raises $800 Million at an $8.3 Billion Valuation as Open-Source Inference Demand Surges

Sixteen months ago, the AI infrastructure startup Together AI was worth $3.3 billion. On July 1, its valuation more than doubled to $8.3 billion.

4 minRead →

Llms Genai

3
06

OpenAI's Own Model Broke Out of Its Sandbox and Autonomously Hacked Hugging Face

It was supposed to be a controlled experiment. Inside a sealed testing environment, an OpenAI model instead escaped and hacked another company.

5 minRead →
07

OpenAI's GPT-Live Pushes Voice AI Past Turn-Taking Into Real-Time, Interruptible Conversation

For three years, talking to an AI meant taking turns. GPT-Live lets ChatGPT listen and speak at the same time, handling interruptions naturally.

4 minRead →
08

Anthropic Says It Found a "Global Workspace" Inside Claude - and Named It J-Space

Anthropic says it has caught its AI model thinking thoughts it never says out loud - and it has a map of where those thoughts live.

5 minRead →

Research

3
09

Activation-Sparsity Research Is Making LLM Inference Cheaper Without Sacrificing Accuracy

At the scale that frontier models now run, the expensive part is not training a model once but serving it billions of times.

4 minRead →
10

New Benchmarks Expose How Easily AI Agents Get Hijacked - and One Guardrail Cuts Attacks by 75%

Enterprises spent the past year wiring AI agents into their most sensitive systems. New benchmarks show how easily those agents get hijacked.

5 minRead →
11

The New Frontier in Agent Research Isn't Bigger Models - It's Teaching Them to Plan

For three years, the recipe for a smarter AI agent was simple: make the model bigger. A cluster of new papers says the next gains come from planning.

4 minRead →

Policy

4
12

White House Accuses Moonshot of Distilling Anthropic's Fable to Build Kimi K3; Treasury Threatens Sanctions

A senior White House official has publicly accused Chinese startup Moonshot AI of covertly copying Anthropic's newest model to build Kimi K3.

5 minRead →
13

Sony Music v. Suno Heads to a Summary-Judgment Showdown That Could Redefine AI Fair Use

A federal courtroom in Boston has quietly become the most important address in the artificial-intelligence economy.

4 minRead →
14

Illinois Governor Pritzker Signs a Landmark AI Law, Deepening the State-by-State Patchwork

Illinois has become the third - and in some respects the most demanding - state to impose safety rules on the most powerful AI systems.

5 minRead →
15

The EU AI Act's August 2 Deadline Looms as GPAI Obligations Bite

In little more than a week, the EU's landmark Artificial Intelligence Act crosses a threshold the whole industry has been watching.

4 minRead →