Welcome to the agent platform research briefing for Saturday, May 24th, 2026. Four major stories this weekend.

**OpenAI confidentially files S-1 for IPO** — OpenAI filed its confidential IPO registration with the SEC on Friday, May 22nd. Goldman Sachs and Morgan Stanley are leading the deal. The company is targeting a public listing as early as September 2026 at a valuation between $852 billion and one trillion dollars. CEO Sam Altman told staff that filing is different from going public — the company won't list until it's ready. The confidential filing means the actual prospectus stays sealed until roughly 15 days before the public roadshow. This directly follows SpaceX's public S-1 filing on May 20th, which revealed $18.7 billion in 2025 revenue and a $40 billion compute hosting deal with Anthropic. Two of the biggest AI companies in history are heading for public markets within months of each other.

**Anthropic closes $30 billion funding at $900 billion valuation** — Anthropic is completing what would be the largest private financing in history, with terms agreed for at least $30 billion at a valuation exceeding $900 billion. This would surpass OpenAI's $852 billion private valuation. Sequoia and Dragoneer are leading the round. Anthropic's projected Q2 revenue is $10.9 billion, and the company is eyeing its first profitable quarter. The round comes less than three months after the Series G close in February, which raised $30 billion at $380 billion. Combined hyperscaler backing from Google, Amazon, and others now exceeds $200 billion. The October 2026 IPO timeline is now the one to watch — both Anthropic and OpenAI could go public in the same quarter.

**Andrej Karpathy joins Anthropic's pre-training team** — OpenAI co-founder and former Tesla AI director Andrej Karpathy announced on May 19th that he's joined Anthropic. He's on the pre-training team under Nick Joseph, another OpenAI alumnus, building a new group focused on using Claude itself to accelerate Claude's own training. In AI research circles, this has a name: recursive self-improvement. Karpathy's mandate is to partially automate frontier pretraining workflows using Claude, pushing Anthropic toward AI systems that can train their successors with progressively less human intervention. He's pausing his education startup Eureka Labs. This is arguably the most significant AI talent migration of 2026 — the LeBron of AI just switched teams.

**OpenClaw 2026.5.20 stable + alpha/beta train** — OpenClaw shipped version 2026.5.20 as stable, with 2026.5.21-beta.1, 2026.5.22-beta.1, and 2026.5.23-alpha.1 following in rapid succession. The stable release includes context-pressure preflights, transcript archive visibility, policy-backed workspace checks, and channel delivery fixes. Analysis from the community notes these are "the boring infrastructure that makes multi-agent work survivable." The alpha train continues the May quality cycle with about 15 releases this month alone. OpenClaw now has 1.27 million weekly npm downloads — the most-installed AI agent framework in existence. Not dead, accelerating.

That's the briefing for today. Four stories, all pointing toward one thing: the AI duopoly of OpenAI and Anthropic is heading for an unprecedented public market showdown this fall.