The AI Brief

OpenAI submitted a confidential S-1 to the SEC on June 8 and announced it proactively, writing: "We recently submitted a confidential S-1. We expect it to leak so we're just announcing it." The company added it has not committed to timing, noting that some strategic objectives "are likely easier as a private company" but that the filing gives it the option to list sooner if conditions allow.

The filing was made at an $852 billion post-money valuation with no confirmed IPO date. Goldman Sachs and Morgan Stanley are lined up as underwriters, with a September 2026 debut reported but not confirmed by OpenAI. The move follows Anthropic's own confidential S-1 on June 1 and a May 2026 jury ruling clearing OpenAI's Public Benefit Corporation restructuring. SpaceX has moved furthest along among the three, having kicked off an investor roadshow ahead of a planned June 12 Nasdaq debut targeting a raise of $75 billion at a valuation of around $1.77 trillion.

Public markets are being asked to price an entity generating $25 billion in annualized revenue while running a negative 122% non-GAAP operating margin, a pairing without precedent in major tech IPO history. Investment bankers have advised both OpenAI and Anthropic that an early-mover advantage is at stake: the first to list would set the terms for how investors categorize the AI sector and gain access to capital seeking an entry point into the industry. The public S-1, once filed, will disclose audited financials, detailed risk factors, and ownership structure, representing the first time investors can assess OpenAI's books with regulatory-grade transparency.

In June 2026, ChatGPT crossed 1 billion global monthly active users, becoming the fastest app in history to hit the milestone, according to Sensor Tower estimates reported by Reuters. ChatGPT reached the mark in roughly three years after launch, outpacing the early growth of Google Maps, TikTok, Instagram, and YouTube. The weekly active user count stands at 900 million, alongside 50 million paying subscribers.

OpenAI is generating $25 billion in annualized revenue as of early 2026, on a roughly $2 billion-per-month run-rate, targeting $29.4 billion for the full year. Menlo Ventures estimates the enterprise LLM market is now concentrated around three leaders: Anthropic at 40%, OpenAI at 27%, and Google at 21%. The consumer-enterprise split has become one of the defining dynamics of the AI platform market: ChatGPT leads on reach, while Claude has taken the enterprise coding and agentic workflow segment.

Caveat: the 900 million weekly active user figure was confirmed by OpenAI in February 2026; the 1 billion monthly active figure is a Sensor Tower estimate reported via Reuters, not a primary OpenAI disclosure. Monthly actives and weekly actives measure different cohorts and are not directly comparable. The revenue and margin figures come from secondary reporting, not audited financials, which will not be available until the public S-1 is filed.

Anthropic released Foundation Models framework support for Claude through a new Swift package. Apple's Foundation Models framework gives developers access to tap into models natively from Swift, returning typed Swift values through guided generation in as few as three lines of code. Developers can use Apple's on-device models for fast, local tasks like summarization or extraction, then hand off to Claude when a request calls for multi-step reasoning, code generation, and more.

The integration works across iOS 27, iPadOS 27, macOS 27, visionOS 27, and watchOS 27. Developers add the Swift package, sign in with an Anthropic API key, and pass typed outputs from Apple's on-device model into a Claude request; the package handles streaming, tool calls, and structured responses back into the SwiftUI view. The architecture means developers can prototype with the on-device model and route to Claude for complex queries within the same SDK and session logic, without changing provider.

Within Apple's broader WWDC 2026 framework, only heavier world-knowledge and complex reasoning requests are routed to the cloud; on-device tasks run on Apple's own foundation models on Apple Silicon. Users can still invoke ChatGPT for specific queries, as since 2024, but Gemini is now the default cloud backbone for Siri. Claude's Swift package integration sits alongside this, available to third-party app developers as a distinct routing option for complex workflows.

Disclosure: Anthropic, mentioned in this item, is the company that develops Claude, which generates this brief.

Governor Polis signed SB 26-189 on May 14, 2026, repealing and replacing SB 24-205 (the original Colorado AI Act). The replacement was the product of a governor-appointed AI Policy Work Group whose proposed framework was released on March 17, 2026. The bill moved with extraordinary speed through the legislature and takes effect January 1, 2027, with enforcement contingent on the attorney general completing rulemaking.

SB 26-189 replaces the original law's comprehensive risk-management framework with a narrower notice-and-transparency model. Risk management programs, annual impact assessments, and the duty to use reasonable care to avoid algorithmic discrimination are removed. What remains includes consumer notice before AI-assisted decisions, adverse-decision notices with appeal rights, and developer documentation requirements.

A separate federal court layer complicates even the replacement law's timeline. On April 27, 2026, a federal magistrate judge stayed enforcement of the original law after xAI challenged it. The stay technically extends to successor legislation, meaning SB 26-189 also cannot be enforced until the injunction question is resolved and rulemaking is complete. Attorney General Phil Weiser has indicated the state will not pursue enforcement until required rulemaking is complete, and that process has not formally begun.

Google DeepMind's AlphaProof Nexus autonomously solved 9 of 353 open Erdős problems and proved 44 OEIS conjectures. The formal proofs are machine-verified in Lean 4. The paper was published on arXiv (2605.22763v1) on May 21, 2026. CEO Demis Hassabis was quick to note the system is "still not AGI."

The system resolved 9 of 353 open Erdős problems it attempted, including two questions that had gone unanswered for 56 years. It also proved 44 of 492 open OEIS conjectures, settled a 15-year-old question about Hilbert functions in algebraic geometry, and improved a known bound in convex optimization. Inference costs ran just a few hundred dollars per problem.

AlphaProof Nexus pairs a frontier LLM with Lean: the model generates candidate proof strategies, Lean accepts or rejects each step. The system uses what DeepMind calls "agentic loops," iterating and refining proofs against the formal checker until they either pass or the system concludes it cannot solve the problem. Unlike pure natural-language proof approaches from other labs, the underlying model does not have to carry the entire logical chain on its own. All proofs are published on GitHub and compile.