OpenAI Model Autonomously Disproves 80-Year-Old Geometry Conjecture
In what Fields medalist Tim Gowers called "a milestone in AI mathematics," an OpenAI internal reasoning model has independently disproved a central conjecture in discrete geometry — the Erdős planar unit distance conjecture — that has stood unsolved since 1946.
What makes this achievement extraordinary is the level of autonomy involved. The model was not fine-tuned for this specific problem, did not retrieve an existing solution, and was not guided step-by-step by human researchers. It received the problem statement and produced the proof independently, discovering an infinite family of constructions using deep algebraic number theory — specifically Golod-Shafarevich theory and infinite class field towers — that achieve polynomial improvement over the traditional square grid approach.
The proof has been verified by a group of external mathematicians, who have also written a companion paper explaining the argument and providing context for the significance of the result. This marks the first time a prominent open problem, central to a subfield of mathematics, has been solved autonomously by AI — a qualitative leap from earlier AI-assisted proofs where humans provided heavy guidance.
Source: OpenAI Blog
Trump White House Scraps AI Oversight Executive Order After Tech Industry Pressure
President Donald Trump postponed — and effectively shelved — a planned executive order that would have established a voluntary framework for AI companies to share advanced models with the government before public release. The framework under discussion would have given federal agencies up to 90 days for security review of new models.
"I didn't like certain aspects of it," Trump told reporters in the Oval Office. "We're leading China, we're leading everybody, and I didn't want to do anything to get in the way of that lead."
The reversal came after direct calls from Elon Musk, Mark Zuckerberg, and former US AI and crypto czar David Sacks, who reportedly warned the president that a government review process could slow American AI development and hand an advantage to foreign competitors. The move comes amid growing concern over the national security implications of Anthropic's Mythos model, which demonstrated advanced cyber vulnerability exploitation capabilities.
The about-face is notable because it represents a policy direction the Trump administration had previously rejected when proposed during the Biden era — only to briefly embrace it before retreating again.
Andrej Karpathy Joins Anthropic to Lead New Pre-Training Research Team
Andrej Karpathy, one of the most recognized names in AI research and a founding member of OpenAI, has joined rival lab Anthropic. The move, announced on May 19, sent shockwaves through the AI community.
Karpathy will work on pre-training under team lead Nick Joseph — the most compute-intensive and expensive phase of building a frontier model. Crucially, he is starting a new team focused on using Claude to accelerate pre-training research itself, an increasingly important approach as AI labs explore recursive self-improvement in their research workflows.
"I think the next few years at the frontier of LLMs will be especially formative," Karpathy wrote on X. "I am very excited to join the team here and get back to R&D."
The hire comes at a pivotal moment for Anthropic, which has been on a tear financially — the company is projecting $10.9 billion in revenue for Q2 2026, up 130% from $4.8 billion in Q1, and is reportedly approaching a $900 billion valuation as it prepares for a potential IPO. Separately, Anthropic also announced a $200 million partnership with the Gates Foundation to develop AI tools for healthcare, education, and agriculture in underserved regions.
Sources: TechCrunch, CNBC
EU Agrees to Simplify AI Act Rules, Delays Key Deadlines
The European Union reached a provisional agreement on the Digital Omnibus package, marking the first set of amendments to the EU AI Act since its adoption in June 2024. The deal, brokered between the Council, Parliament, and Commission, delays and simplifies several key obligations.
High-risk AI system obligations under Annex III have been postponed from August 2026 to December 2027 — a 16-month extension — while product-regulated high-risk systems get a one-year delay to August 2028. The amendments also extend regulatory exemptions previously granted to SMEs to cover small mid-cap companies, and reduce governance fragmentation by clarifying the competences of the AI Office.
Notably, the agreement introduces new prohibitions: AI systems that generate non-consensual intimate imagery of identifiable persons and AI-generated child sexual abuse material will be explicitly banned, effective December 2026.
Formal adoption is expected in June with publication in July. The simplification is widely seen as a pragmatic response to industry feedback that the original timeline was too aggressive, particularly given that many technical standards are still being developed.
Source: Council of the EU
Quick Hits
Google I/O 2026: Google unveiled Gemini Spark, a 24/7 personal AI agent, along with the Gemini Omni world model and Gemini 3.5 Flash. The AI Ultra subscription dropped from $250/month to $200, with a new $99 tier. CNBC
Meta controversy: Leaked audio of Mark Zuckerberg revealed that Meta had been tracking employees' Gmail, coding sessions, and internal tools to train AI — disclosed on the same day 8,000 employees were fired.
OpenAI IPO: OpenAI has filed for an initial public offering, and a SpaceX S-1 filing revealed a $1.25 billion/month deal with Anthropic for AI services.
Anthropic creative tools: Anthropic introduced connectors allowing Claude to integrate directly with Adobe apps, Blender, and Ableton, expanding its reach into creative workflows.