AI Weekly Roundup (May 25‑31 2026): OpenAI Solves 80‑Year Math Problem, Anthropic Hits $109B Revenue and $300B Funding, Plus Major Industry Moves

OpenAI proves the Erdős unit‑distance conjecture

OpenAI’s internal general‑reasoning model independently demonstrated a proof of the Erdős unit‑distance conjecture, a core problem in discrete geometry posed in 1946. The proof spans 125 pages and relies on algebraic number theory and the Golod‑Shafarevich criterion. Fields Medalist Tim Gowers and Princeton mathematician Will Sawin verified the result, calling it a milestone for AI mathematics.

Anthropic posts $109 billion Q2 revenue and first quarterly profit

According to the Wall Street Journal and TechCrunch, Anthropic’s second‑quarter revenue reached roughly $109 billion, generating an operating profit of $5.59 billion and an annualized revenue run‑rate of $440 billion. High‑spending customers (>$100 k annual spend) grew seven‑fold, and the number of million‑dollar clients rose from 12 to over 500.

Anthropic’s $300 billion financing round could eclipse OpenAI

Bloomberg and the Financial Times report that Anthropic is close to closing a financing round exceeding $300 billion, valuing the company at over $900 billion—potentially surpassing OpenAI’s $8.52 trillion valuation and making Anthropic the world’s highest‑valued private AI firm.

OpenAI files a confidential S‑1 for a September 2026 IPO

CNBC and Fortune cite that OpenAI submitted a confidential S‑1 filing on May 22, targeting a September 2026 IPO. The filing projects a valuation between $8.52 trillion and $10 trillion, with the confidential filing to public IPO transition expected to take 4‑6 months. This occurs amid a broader $3.7 trillion AI IPO wave that includes SpaceX and Anthropic.

U.S. President cancels AI safety executive order after calls from tech CEOs

President Trump withdrew a planned AI safety executive order after direct calls from Elon Musk, Mark Zuckerberg, and David Sacks. The administration’s security team had spent weeks building a regulatory framework before the order was scrapped.

Pope Leo XIV releases the first AI encyclical, “Magnifica Humanitas”

On May 25, Pope Leo XIV issued the inaugural AI‑focused encyclical, addressing AI’s impact on human dignity, the replacement of creative work, labor rights in the automation era, concentration of AI power, and risks such as deception and autonomous weapons.

Andrej Karpathy joins Anthropic to rebuild its pre‑training research team

Former OpenAI co‑founder and ex‑Tesla Autopilot lead Andrej Karpathy announced his move to Anthropic, stating that the next few years of LLM development will be especially formative.

Google Gemini integrates Canva, Adobe, and CapCut to become a full‑stack creative studio

At Google I/O 2026, Gemini announced native integrations: Canva’s “Magic Layers” for editable image layers, Adobe’s Firefly AI Assistant for complex workflows in Photoshop and Premiere (upcoming), and CapCut’s mobile video‑editing features (upcoming).

Google Search undergoes its biggest redesign in 25 years

Google is rebuilding the search bar itself, enabling on‑page generation of custom visuals, interactive graphics, and mini‑apps directly from queries. VP Robby Stein noted that user questions are becoming longer and harder, with no clear answers online.

Intuit cuts 17% of its workforce and pivots to AI agents

Intuit announced a 17% layoff (≈3,000 employees) while shifting focus to AI agents, operational efficiency, and mid‑market services, projecting over $500 million in annual cost savings for the second half of 2026.

Anthropic acquires Stainless to strengthen its API SDK tooling

Anthropic confirmed the purchase of Stainless, a startup that builds polished SDKs for APIs such as OpenAI, Cloudflare, and Merge, supporting Python, TypeScript, Java, Go, and Ruby.

OpenAI revokes its macOS app signing certificate, triggering a supply‑chain attack

On June 12, OpenAI will fully revoke its macOS code‑signing certificate, leading to the TeamPCP supply‑chain breach. Forensic analysis revealed two employee devices were compromised, with limited credential material stolen from internal repositories; iOS and Windows certificates have been rotated.

Tencent open‑sources TencentDB Agent Memory, a 4‑layer local memory architecture for AI agents

Tencent released the MIT‑licensed TencentDB Agent Memory, a fully local, four‑layer memory system that eliminates the need for cloud dependencies and solves persistent, structured‑memory challenges in production deployments.

Perplexity open‑sources Bumblebee, a read‑only supply‑chain security scanner

Perplexity launched Bumblebee, a scanner that inspects npm, PyPI, Go modules, MCP configurations, and browser extensions, addressing a real and underserved gap in AI‑development environment security.

ByteDance raises its 2026 AI infrastructure budget by 25% to ¥2 trillion

ByteDance increased its AI capital‑expenditure budget to ¥2 trillion, allocating ¥850 billion to AI chips and over $5 billion to domestic chip pre‑purchases (e.g., Cambricon, Huawei Ascend).

Baidu unveils Wenxin Large Model 5.1 with “multidimensional elastic pre‑training”

Baidu’s new foundation model, Wenxin 5.1, claims pre‑training costs at roughly 6% of industry‑standard models while achieving top rankings on the LMArena search leaderboard in China.

White House drafts an AI model review executive order modeled on FDA drug approval

Kevin Hassett, director of the National Economic Council, disclosed that the administration is preparing an executive order to subject new AI models to a review process analogous to FDA drug approvals.

Five‑Eyes alliance issues joint guidelines on deploying agentic AI safely

The alliance warned of unique risks such as productivity loss, service disruption, privacy breaches, and cyber incidents, recommending low‑risk task limits, identity‑management controls, human‑in‑the‑loop supervision for high‑cost operations, and regular red‑team testing.