AI & Tech · Week 44, 2026
AI/Tech this week: OpenAI's maths claims, Gemini 4 Carbon, Anthropic files to go public
OpenAI put AI-generated maths and physics results in front of academics, Anthropic filed the paperwork for a stock market listing, and two new models arrived with cyber security as the headline feature.
- 1Research
OpenAI publishes model-generated maths and physics results for academics to check
OpenAI shared results its models produced on open problems in maths and physics, including the Kakeya conjecture, the Exponential Time Hypothesis and quantum chromodynamics. One preprint, posted in a public GitHub repository, is titled "The Exponential Time Hypothesis is False". OpenAI set out how researchers can verify the work and build on it.
Why it matters: If independently confirmed, this would be one of the clearest signs yet that AI systems can contribute original results to serious research.
What's next: OpenAI has published the preprints so academic communities can attempt to verify the claims and build on them.
Learn the basics: Reasoning model →
Sources: openai.com, github.com
- 2Models
Google announces Gemini 4 Carbon, aimed at long tasks and cyber defence
Google announced Gemini 4 Carbon, which it says ranks first on version 5.0 of the Artificial Analysis Intelligence Index with a score of 59 and leads benchmarks for finding, exploiting and patching software flaws. Pricing starts at $2 per million input tokens and $10 per million output tokens.
Why it matters: Models that can find and fix software flaws change day-to-day security work, which is why Google is handing this one to vetted defenders first.
What's next: Access begins with a subset of cyber defenders, widening to trusted partners through Google's Fairwind Program before a broader developer and consumer release.
Learn the basics: Frontier model →
Source: blog.google
- 3Funding
Anthropic files S-1 paperwork for a stock market listing
Anthropic filed an S-1 registration statement with the US Securities and Exchange Commission on 26 October, the document that precedes an initial public offering. It has applied to list Class A shares on the Nasdaq Global Select Market and remains a public benefit corporation, with a Long-Term Benefit Trust that elects a majority of its board.
Why it matters: A listing would open the Claude maker's finances to public scrutiny and let ordinary investors buy shares in a leading AI lab.
What's next: The filing leaves the share price range and number of shares blank; both must be set before the offering can proceed.
Learn the basics: AI unicorns →
- 4Models
Mistral releases the weights of Mistral Large 4, its trillion-parameter model
French lab Mistral released the weights of Mistral Large 4, a 1-trillion-parameter multimodal model, so anyone can download, customise and self-host it. Mistral says it leads open-weight models on coding benchmarks including FrontierCode and Terminal-Bench, scores 51% on a cyber defence index, and was trained from scratch on European infrastructure.
Why it matters: Open weights let companies and governments run a capable model on their own machines rather than sending data to someone else's service.
Learn the basics: Open-weight model →
Source: mistral.ai
- 5Other
Backdoor found in Claude Code extension on the Cursor Marketplace
Security firm Phorion found a backdoor in the official Claude Code for VS Code extension bundle distributed on the Cursor Marketplace. Versions 2.1.312 and 2.1.313 added an attacker's SSH key to the user's machine and tried to switch on remote access across macOS, Linux and Windows. Cursor removed the affected versions.
Why it matters: Coding extensions run with your own permissions, so one tampered update can expose source code and developer tools across an entire company.
What's next: Phorion says organisations should check whether those versions were installed and engage incident response where they are found.
Learn the basics: Supply-chain attack →
Sources: phorion.io, doublepulsar.com
Term of the week
Supply-chain attack
An attack that tampers with software you already trust, such as an update, library or editor extension, so that installing it also installs the attacker's code.
Read the full explanation →