Tuesday, 08 September 2026

Mathematician alleges OpenAI misappropriated his Navier-Stokes proof; Mistral raises €3B in Europe's largest-ever tech equity round; and 28x more AI spend fails to move the innovation needle.

Today's Lead

Engineering

Tristan Buckmaster / NYU Courant Institute

Statement on Finite-Time Blowup for Forced 3D Euler, and OpenAI's Conduct

Mathematician Tristan Buckmaster published a statement accusing OpenAI of misconduct after OpenAI announced a proof of the Navier-Stokes Millennium Problem days after learning of Buckmaster's parallel work. Buckmaster alleges OpenAI began their effort only after receiving information about his research, initially misrepresented the level of human involvement in their proof, and declined to answer whether his private Codex chat sessions were used in training data. An OpenAI representative reportedly made a veiled threat when Buckmaster objected to the company's publication terms.

Read →

Engineering

Mistral AI

Mistral Makes Sovereign Open-Weight AI to Frontier

Mistral raises €3 billion in funding, marking the largest equity fundraising round by a European technology company. Samsung Electronics leads the investment, with support from Scaleup Europe Fund, PSG Equity, and many others. Mistral plans to expand frontier research, scale compute capacity for training powerful models, and accelerate commercial growth. The company aims to build sovereign AI that lets organizations control their data, use customizable models, and run auditable systems independent of single vendors.

Read →

mcpherrin.ca

I've Factored the RSA Keys of a Certificate Authority from the 90s

The author factored 512-bit RSA keys from 1990s Canadian certificate authorities using the CADO-NFS algorithm. A desktop computer took 29 to 32 hours per key, while a collaborator used a GPU cluster to factor another key in one hour. These keys were too weak even in 1999, and this work shows why cryptographic standards must update to address quantum computing threats.

Read →

Dan Luu

How Well Do Agents Use Test/Verification Techniques?

AI agents perform poorly with most testing methods, even when instructed to use them. Default instructions work better than specialized techniques like formal methods or property-based testing. Agents struggle to implement these techniques correctly and generate trivial tests instead. Iterative feedback with human guidance produces better results than standalone techniques alone.

Read →

LeadDev

28x More AI Spend. Zero New Innovation.

AI spending increased 28 times, yet engineering time devoted to new features remained flat. Efficiency gains flow into organizational bottlenecks rather than innovation. Only 6% of executives measure positive AI return on investment. Organizations must track innovation ratios and reinvest freed capacity into discovery work.

Read →
Humanities

JSTOR Daily

The Nazi Car That Conquered the World

Hitler created the Volkswagen as an affordable car called the KdF-Wagen. World War II production diverted from civilian cars to military vehicles. After the war, production restarted and the Beetle became the world's best-selling car. Historian Natalie Scholz argues that Germans deliberately obscured the car's Nazi origins. The Beetle remained a public secret: buyers owned it but did not acknowledge its past. Volkswagen accepted responsibility for this history and created a compensation fund in 1991.

Read →