The Polylog AI Intelligence Brief

Morning Edition · Saturday, August 29, 2026

Tech

Z.ai Opens the Weights of GLM-5.3, Its Strongest Coding and Cyber Model

The Chinese lab delayed the download for two weeks after the model scored 84.5% on the CyberGym vulnerability benchmark, a score it says beat Anthropic's Mythos 5.

3 sources
Corroborated

Tech

OpenAI Will Cut Cursor Off From Its Models on November 12 After the SpaceX Takeover

The company says it cannot be confident SpaceX will honor its terms of service, citing what happened after Elon Musk acquired Twitter.

3 sources
Corroborated

Tech

Anthropic Opens a Preview of a Standard That Lets AI Agents Operate Lab Hardware

Early users report a laser stabilization rate on QuEra quantum computers rising from 58% to 99.3% and an imaging experiment compressed from weeks to a day.

3 sources
Corroborated

Tech

Anthropic Says Automated Researchers Fixed Alignment Failures on All Ten Benchmarks It Tried

The automated systems ran at roughly $4 per hour in inference cost against $150 per hour for the human researchers used as the comparison.

3 sources
Corroborated

Tech

IBM Ships Granite 4.2 Reasoning Models From 3B to 30B Under Apache 2.0

The 30-billion-parameter model scores 57.00 on SWE-bench Verified and 89.17 on AIME25, with a switch that turns step-by-step reasoning off for latency-sensitive calls.

3 sources

Tech

Meta Prices Its Muse Spark API Far Below OpenAI and Anthropic Output Rates

Meta charges $1.25 per million input tokens and $4.25 per million output tokens for a model with a one-million-token context window.

3 sources

Tech

Berkeley Lab Cuts Beamline Image Analysis From a Month to Minutes Using Meta's Open Vision Models

The SYNAPS-I project runs Segment Anything 3 and DINOv3 across 300 Nvidia A100 accelerators to segment X-ray and neutron imaging data.

3 sources

Tech

Hugging Face Puts a $399 Bipedal Robot and Its Full Training Stack on Sale

The 25-centimeter Microduck carries 15 motors, a camera, a small lidar and two inertial sensors, with behaviors trained in simulation and deployed directly to the hardware.

3 sources

Tech

OpenAI Is Testing a Codex Mode That Keeps Working Until a User Stops It

Code reviewed by WIRED describes an agent that generates its own follow-up tasks across sessions and can message a user unprompted.

2 sources
Corroborated

Macro

Bill Gates Calls for Taxing AI and Robots and Says There Is No Plan for Displaced Workers

He argues employers pay payroll taxes on workers while deducting machines as business expenses, and proposes reserving a category of jobs for humans.

2 sources
Corroborated

Tech

Anthropic Watermarks Claude Text Using Google DeepMind's SynthID Method

Every Claude model released after August 2 marks its output, and a detected watermark means Claude processed the text, not that Claude wrote it.

3 sources

Tech

OpenAI Opens an Accelerator With Thailand's Science Ministry for Ten Local Startups

Each company receives $2,000 in interface credits over eight weeks, in OpenAI's first public-private partnership with the Thai government.

3 sources
Corroborated