Polylog
The Polylog AI Intelligence Brief

Morning Edition · Monday, July 6, 2026

AI Capital Spending Set to Pass US Defense Budget in 2027, Morgan Stanley Projects

Markets

AI Capital Spending Set to Pass US Defense Budget in 2027, Morgan Stanley Projects

Combined spending by five large cloud companies is forecast to reach about 3.2% of gross domestic product, raising the question of whether end-customer revenue can service the buildout.

3 sources
Corroborated
Anthropic Ships Claude Sonnet 5, Pitching Cheaper Autonomous Agents

Tech

Anthropic Ships Claude Sonnet 5, Pitching Cheaper Autonomous Agents

The mid-tier model scores 92.4% on SWE-bench Verified and costs less than earlier flagship models, continuing a trend in which greater capability becomes available at lower cost.

3 sources
OpenAI Previews GPT-5.6 Sol in Codex, Timed Against Anthropic

Tech

OpenAI Previews GPT-5.6 Sol in Codex, Timed Against Anthropic

Leaked text strings and an official preview point to a Sol, Terra, and Luna family, with a top-tier "Sol Ultra" version leading command-line coding benchmarks.

3 sources
Corroborated
Meta's Brain2Qwerty v2 Decodes Typed Sentences From Brain Scans at 61% Accuracy

Tech

Meta's Brain2Qwerty v2 Decodes Typed Sentences From Brain Scans at 61% Accuracy

A non-invasive magnetoencephalography pipeline approaches the accuracy of surgical decoders, but remains too error-prone for daily use.

2 sources
Anthropic Redeploys Claude Fable 5 With New Cyber Classifiers and a Jailbreak Standard

Tech

Anthropic Redeploys Claude Fable 5 With New Cyber Classifiers and a Jailbreak Standard

After conversations with the United States government, Anthropic reinstated the model worldwide and proposed an industry framework, developed with Amazon, Microsoft, and Google, for scoring the severity of jailbreaks.

3 sources
Corroborated
Anthropic Opens Claude Science, an Agent Workbench for Researchers

Tech

Anthropic Opens Claude Science, an Agent Workbench for Researchers

The app runs on Claude Opus 4.8, includes more than 60 scientific databases and a citation-checking reviewer agent, and is available to paid subscribers without special access requirements.

3 sources
Meta's Muse Spark Points to Compute Efficiency as the New Battleground

Tech

Meta's Muse Spark Points to Compute Efficiency as the New Battleground

The first model from Meta Superintelligence Labs claims frontier capability using more than an order of magnitude less compute than Llama 4 Maverick, and it powers Meta's smart glasses.

3 sources
China's Humanoid Robot Boom Runs Into Hardware and Data Limits

Tech

China's Humanoid Robot Boom Runs Into Hardware and Data Limits

With more than 140 manufacturers and heavy state funding, the industry faces a shortage of training data, fragile robotic hands, and government warnings about a bubble.

3 sources
Corroborated
France Moves to Field Its First AI-Directed Combat Unit by 2027

Geopolitics

France Moves to Field Its First AI-Directed Combat Unit by 2027

The Pendragon program aims to deploy about 15 ground robots and 60 drones for autonomous missions, commanded by a single captain who assigns tasks to the AI.

3 sources
Corroborated
Segment Anything Underpins a Fashion App, Showing Open-Vocabulary Vision in Production

Tech

Segment Anything Underpins a Fashion App, Showing Open-Vocabulary Vision in Production

Alta Daily used Meta's Segment Anything to process more than 20 million clothing images, an example of promptable segmentation moving from research to product.

3 sources
Two Philosophies Harden at the Top of the AI Industry

Tech

Two Philosophies Harden at the Top of the AI Industry

As large amounts of capital enter the field, the public split between a risk-first camp and an acceleration-first camp increasingly shapes how labs raise money and lobby for rules.

3 sources
Plausible
A Claimed Billion-Scale Facade Dataset Surfaces, Awaiting Verification

Tech

A Claimed Billion-Scale Facade Dataset Surfaces, Awaiting Verification

A research post advertises centimeter-accurate, cross-continental building-facade point clouds with fine-grained semantic labels, the kind of spatial data used to train 3D perception models.

2 sources