The Omniscient Bulletin · 2026-05-21
The Omniscient Bulletin — May 21, 2026
OpenAI solves a geometry problem that stumped mathematicians for 80 years, Anthropic eyes its first profit, and Trump pulls back on AI oversight.
Thursday surfaced a productive tension at the heart of the AI moment. An OpenAI reasoning model produced an original proof that dismantled a geometry conjecture Paul Erdős posed in 1946 and no human had cracked since - and did so without being purpose-built for mathematics. Hours later, the White House canceled a signing ceremony for an executive order that would have created a voluntary framework for vetting exactly these kinds of capability advances before public release. Trump's stated reason: he did not want anything getting in the way of America's lead over China. The juxtaposition is telling. Meanwhile, Anthropic is closing in on its first operating profit, the SpaceX IPO filing offers the first public look at xAI's books and the losses are considerable, and a new consumer AI lab just raised $700 million on the bet that the hardware layer remains wide open. A full issue today.
Frontier
OpenAI's reasoning model disproves an Erdős geometry conjecture unsolved since 1946
An OpenAI general-purpose reasoning model has produced an original proof disproving a conjecture posed by Paul Erdős in 1946 - one that led mathematicians to believe optimal point-set configurations resemble square grids. Independent mathematicians including Noga Alon and Thomas Bloom, who maintains the Erdős Problems site and previously rebuked a premature OpenAI claim on the same topic, reviewed and validated the result. OpenAI says it marks the first time AI has autonomously solved a prominent open problem in mathematics - and that the model was not purpose-built for the task.
Industry
Anthropic projects first-ever operating profit on $10.9B Q2 revenue
Anthropic has told investors it expects Q2 revenue to more than double quarter-over-quarter to roughly $10.9 billion, with an operating profit appearing for the first time, per the Wall Street Journal. The profitability window may be brief: large compute commitments later in the year are expected to push the company back into the red. The milestone is notable given Anthropic's $40 billion valuation raise earlier this year - and raises the question of how quickly the margin picture changes once those compute contracts kick in.
Compute
SpaceX IPO filing reveals xAI lost $6.4B in 2025 and plans to scale Grok to trillions of parameters
xAI recorded a $6.4 billion operating loss on $3.2 billion in revenue in 2025, with AI segment capex hitting $7.7 billion in Q1 2026 alone - an annualized rate exceeding $30 billion. SpaceX's IPO filing, the first public look at xAI's books, also discloses plans to scale Grok to 'multiple trillions of parameters' and to begin deploying orbital AI compute satellites as early as 2028. Grok AI features are used by 117 million of the combined platform's 550 million monthly active users.
Policy
Trump pulls back AI executive order at the last minute, citing concern it would slow AI development
President Trump declined to sign a prepared executive order on government oversight and access to AI on Thursday, saying the language 'could have been a blocker' for jobs and AI progress. Trump also cited U.S. competition with China as a reason to avoid new constraints, telling reporters 'we're leading China, we're leading everybody.' The order would have established a voluntary framework for vetting the national security risks of advanced AI systems before public release, with Anthropic, OpenAI, and Google among the named participants.
Industry
Spotify and Universal Music ink AI licensing deal for fan-made covers and remixes
Spotify announced a licensing agreement with Universal Music Group that will let Spotify Premium subscribers use generative AI to create covers and remixes of UMG artists' songs, with a revenue share flowing back to participating artists. Pricing and launch date were not disclosed. The deal, announced at Spotify's Investor Day alongside a raft of AI-powered podcast and audiobook tools, marks a significant shift: where AI music startups like Suno and Udio faced label lawsuits, Spotify went straight to rights holders for a consent-first framework.
Frontier
OpenAI safety executive Aleksander Madry exits to focus on AI's economic impact
Aleksander Madry, who built and led OpenAI's Preparedness team before being reassigned to an AI reasoning role last summer, announced Thursday he is leaving - with a reported September IPO now weeks away. He says his next focus will be AI's economic impact. The departure extends a pattern of safety leadership attrition at OpenAI: former chief scientist Ilya Sutskever and safety team head Jan Leike both exited in 2024, and the Preparedness team has seen repeated reorganization since.
Industry
Hark raises $700M Series A at $6B valuation to build a consumer AI interface and hardware
Hark, the AI lab founded by serial entrepreneur Brett Adcock, raised $700 million in a Series A round led by Parkway Venture Capital and backed by Nvidia, Qualcomm Ventures, AMD Ventures, ARK Invest, and others. The company, which Adcock seeded with $100 million of his own capital in late 2025, is building a multimodal personal AI assistant paired with purpose-built hardware - targeting the consumer AI gap that coding-focused labs like Anthropic and OpenAI are not filling. Hark has 70 employees and expects to ship its first models this summer.