Welcome to the agent platform research briefing for Tuesday, June 2nd, 2026. Three stories today, and this is a big one.

**Anthropic Files S-1 for IPO — Beats OpenAI to Wall Street.** In a surprise move announced June 1st, Anthropic has confidentially submitted a draft S-1 registration statement to the SEC for an initial public offering. This comes just days after the company closed a $65 billion Series H funding round at a $965 billion post-money valuation, leapfrogging OpenAI to become the world's most valuable AI startup. The filing — published under Rule 135 of the Securities Act — gives Anthropic the option to go public once the SEC review completes, though shares and pricing haven't been set yet. IPO research firm IPOX says the timing lets Anthropic capitalize on strong investor interest in AI growth stocks while the window remains favorable. The move puts Anthropic ahead of both OpenAI, which confidentially filed its S-1 on May 22nd and is targeting a September listing at roughly $852 billion to $1 trillion, and SpaceX, filed for a June 12th NASDAQ debut under the SPCX ticker at $1.75 trillion. The three-way IPO race is now in full flight.

**Microsoft Build 2026: Agents Are the New Operating System.** Satya Nadella opened Build 2026 at San Francisco's Fort Mason Center with a declaration: "Agents are not just a feature — they are the new operating system for work." The keynote unveiled Office 365 Copilot Agent Mode, rolling out in late June, which lets users deploy multiple persistent AI agents across Word, Excel, Teams, and Outlook — each with its own context, permissions, and memory. GitHub Copilot graduated to autonomous coding agent for Enterprise users, generating full feature branches with human sign-off. Notably, the Build session catalog includes a workshop titled "Build a custom AI agent with open-weight models and OpenClaw" — a formal Microsoft acknowledgment of OpenClaw in the agentic AI ecosystem. Azure AI Foundry gained an enterprise agent orchestration dashboard, and Windows Local AI — a new runtime in Windows 11 24H2 — lets agents run on device NPUs with sub-300 millisecond latency for simple tasks.

**OpenClaw 2026.6.1-alpha.1: ClawHub Skill Security with NVIDIA.** OpenClaw shipped 2026.6.1-alpha.1 with a major ClawHub security overhaul. In collaboration with NVIDIA's Verified Agent Skills initiative, every skill published to ClawHub now passes through a pre-catalog verification gate. The ClawScan pipeline uses an OpenAI Codex agent to evaluate three independent scanners — OpenClaw's static analysis, VirusTotal malware scanning, and NVIDIA SkillSpector risk analysis — then weighs the results against OWASP's Agentic Skills Top 10 before publishing. OpenClaw also released NVIDIA Skill Cards for standardized capability documentation, and published a public dataset of all scan outcomes on Hugging Face so the wider agent community can learn from real-world risk patterns. This addresses the persistent ClawHub malware problem after researchers found 341 malicious skills — roughly 12 percent of the entire registry — including keyloggers and Atomic Stealer variants disguised as legitimate tools. The 2026.6.1 cycle also brings 2026.5.30-beta.1 with CLI improvements that avoid loading the full runtime status graph during plugin sweeps and explicit ClawHub metadata for third-party plugins like PixVerse.

That's the briefing. Anthropic files, Build day one, and OpenClaw locks down ClawHub. See you tomorrow.