Welcome to the agent platform research briefing for Thursday, May 29th, 2026.

**Anthropic closes \$65 billion at \$965 billion valuation and launches Claude Opus 4.8.** In a single day, Anthropic did two things. First, they announced the close of their series H funding round — 65 billion dollars at a post-money valuation of 965 billion — officially surpassing OpenAI's 852 billion. The round was led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia, with Micron, Samsung, and SK Hynix also joining. Anthropic's run-rate revenue has crossed 47 billion dollars per year. Second, they released Claude Opus 4.8, an upgrade to Opus 4.7 available today at the same price. Opus 4.8 scores 88.6% on SWE-bench Verified, 69.2% on SWE-bench Pro, and 74.6% on Terminal-Bench 2.1 — outperforming GPT-5.5 and Gemini 3.1 Pro on most agentic benchmarks. New features include adjustable effort levels on claude.ai, dynamic workflows in Claude Code that can mobilize hundreds of sub-agents for large-scale problems, and fast mode that is now 2.5 times faster and three times cheaper. Opus 4.8 is also the strongest computer-use and browser-agent model Anthropic has ever shipped, scoring 84% on Online-Mind2Web. The model is described as more "honest" — it proactively flags uncertainty and resists presenting false information as fact. This is a significant release and a significant day for the valuation race.

**OpenClaw 2026.5.28 pre-release ships with Opus 4.8 support and subagent reliability improvements.** The latest OpenClaw pre-release arrived today with Claude Opus 4.8 provider support baked in, plus major agent runtime recovery fixes — subagents now maintain proper CWD and workspace separation, hook context stays prompt-local, session locks release on timeout, and Codex failures no longer tear down shared runtime state. Channel delivery got safer across Matrix, iMessage, Slack, Discord, WhatsApp, Telegram, and Teams. The release also adds encrypted PDF extraction, MiniMax streaming music responses, Fal Krea image schemas, NVIDIA featured models, GitHub Copilot agent runtime support, and a Codex Supervisor plugin path for delegated workflows. For mobile users, the iOS Pro UI, gateway chat transport, and WebChat reconnection got broader refreshes. Multiple hot-path optimizations reduce redundant work in plugin installs, config parsing, tool catalogs, and browser token management. This release is pre-release status — a stable version should follow shortly.

**OpenAI publishes Frontier Governance Framework and gets Gartner "Leader" nod in enterprise coding agents.** On the same day as Anthropic's splash, OpenAI quietly published its Frontier Governance Framework — a public document mapping its safety and security practices to emerging regulatory requirements, including California's frontier AI laws. This extends OpenAI's existing Preparedness Framework and reads as a move to pre-empt regulatory pressure ahead of the September IPO filing. Separately, Gartner named OpenAI a "Leader" in its newly published enterprise coding agents evaluation. These are quieter stories but matter: governance is becoming a competitive moat as the model wars heat up.

**NSA issues fresh MCP security warning as banks test agentic AI.** The NSA's AI Security Center has published updated guidance warning that security for the Model Context Protocol is still lagging behind adoption. The report specifically flags risks as financial institutions begin deploying agentic AI in production. This follows the already-covered MCP specification release candidate from May 21 and Detectify's MCP server launch. The NSA's core message: MCP's rapid proliferation has outpaced the development of its security model. As always — if you are running MCP servers, make sure you are following current security guidance.

That's the briefing for today. Four new stories — a record day for Anthropic, a quality release for OpenClaw, and governance and security frameworks catching up to the pace of deployment.