AI news you can't miss this week

OpenAI's model cracks an 80-year-old math conjecture, Anthropic hires OpenAI co-founder Andrej Karpathy, Anthropic taps Microsoft's Maia chips to scale AI, OpenAI beats Musk's $100B lawsuit in court & ChatGPT connects to 12,000+ banks via Plaid

Best AI news of this week:

1️⃣ OpenAI's reasoning model autonomously disproved a 1946 Erdős conjecture — the first AI-driven discovery of a major open math result.

2️⃣ Andrej Karpathy, OpenAI co-founder and ex-Tesla, joins Anthropic's pre-training team.

3️⃣ Microsoft is in talks to supply its custom Maia chips to Anthropic, easing its compute crunch.

A jury threw out Elon Musk's $100B lawsuit against OpenAI, ruling every claim was filed years too late.

ChatGPT now links to 12,000+ banks via Plaid, giving Pro users interactive spending dashboards in a US preview.

WEEKLY AI RECAP

MAY 18th - 23rd 2026

🧮 OpenAI's model breaks an 80-year-old math conjecture 🧮
A general-purpose reasoning model achieves a milestone in autonomous mathematical discovery.

OpenAI announced that an internal reasoning model has autonomously disproved a long-held belief related to Paul Erdős' 1946 unit distance problem — what the company calls the first AI-driven discovery of a major open math result. For decades, mathematicians assumed optimal solutions followed grid-like constructions, but the model uncovered an entirely new family of structures using algebraic number theory. The proof was verified by prominent mathematicians including Fields Medalist Tim Gowers, Noga Alon, and Thomas Bloom. OpenAI stressed this came from a general reasoning model, not a specialized system like AlphaProof. The claim follows an earlier retraction where GPT-5 had reportedly solved Erdős problems already present in academic literature — a strong competitive signal for AI-driven scientific discovery.

🧠 OpenAI co-founder Andrej Karpathy joins Anthropic 🧠
One of the most recognized figures in modern AI moves to a rival frontier lab.

Andrej Karpathy, one of the most recognized figures in modern AI research, has officially joined Anthropic. A co-founder of OpenAI in 2015, he later led Tesla's Autopilot vision team from 2017 to 2022, briefly returned to OpenAI, then left in 2024 to launch Eureka Labs, an AI-native education startup, while becoming one of the most followed AI educators online through his "Neural Networks: Zero to Hero" lectures. Now he joins the pre-training team under Nick Joseph and will lead a new internal effort to apply Claude to Anthropic's own training pipeline. Calling the next few years at the LLM frontier "especially formative," his move marks a major competitive shift among leading labs and strengthens Anthropic's research bench as the race for frontier capabilities intensifies.

⚡️ Anthropic taps Microsoft's Maia chips to scale AI ⚡️
Microsoft's custom silicon could replace part of Anthropic's Nvidia compute dependency.

Microsoft is in discussions to supply its custom Maia AI chips to Anthropic, which continues to face significant compute constraints despite existing cloud partnerships with Amazon and Google. The Maia chip reportedly delivers 30% improved performance, making it an attractive option as Anthropic scales its AI-assisted programming workloads. This potential collaboration follows Microsoft's $5 billion investment in Anthropic last November, further deepening the strategic ties between the two companies. If finalized, it would position Microsoft's custom silicon as a credible alternative to Nvidia's dominance — a competitive shift that could reshape the AI chip supply landscape, while strengthening Microsoft's infrastructure strategy beyond its own Azure cloud services.

INTERESTING TO KNOW

⚖️ Jury dismisses Musk's $100B lawsuit against OpenAI ⚖️

A jury unanimously dismissed Elon Musk's lawsuit seeking over $100 billion from OpenAI, Sam Altman, Greg Brockman, and Microsoft, ruling all claims were filed years too late. Musk had alleged that Altman and Brockman "stole a charity" by converting OpenAI into a for-profit entity. OpenAI's legal team countered that Musk himself had supported a for-profit structure and only sued after launching his rival xAI in 2023. Musk called the outcome a "calendar technicality" and vowed to appeal. The dismissal clears a key legal obstacle for OpenAI's ongoing corporate restructuring, letting the company move forward without the overhang.

🏦 ChatGPT connects to 12,000+ banks via Plaid 🏦

OpenAI has launched a preview of a personal finance experience inside ChatGPT, initially available to Pro subscribers in the US. The feature lets users securely connect bank accounts, brokerages, and credit cards through Plaid, with support for Chase, Schwab, Robinhood, and over 12,000 other institutions. Once linked, users get interactive spending dashboards and can ask questions grounded in their real financial data, with Intuit integration planned next. Critically, ChatGPT can analyze but cannot move money, make trades, pay bills, or file taxes — putting real competitive pressure on budgeting apps like Mint and YNAB.

📩 Have questions or feedback? Just reply to this email , we’d love to hear from you!

🔗 Stay connected: