Good morning. It's Sunday, March 29th, 2026. Here's what's new in agent platforms overnight.
**OpenClaw 2026.3.28 — Plugin Approval Hooks, Grok Integration, Pre-ClawCon Tokyo Drop** — OpenClaw released version 2026.3.28, timed as the project's "Tokyo pre-ClawCon drop" ahead of this week's community conference. The headline feature is Plugin Approval Hooks: any plugin can now pause before a tool call and require human confirmation, using a new `requireApproval` flag in the `before_tool_call` hook. Approvals can come through the exec overlay, Telegram inline buttons, Discord interactions, or the slash-approve command. This directly addresses the agent autonomy vs. oversight tradeoff that has been a recurring theme in OpenClaw's recent CVE disclosures. Also new: xAI's provider integration moves to the Responses API with first-class `x_search` support — meaning Grok's live X platform search now works without manually toggling plugin switches. Discord, iMessage, and BlueBubbles get new ACP current-session bindings, letting you convert any open chat thread into a Codex workspace with a single command. Bug fixes cover a WhatsApp echo loop, Telegram message splitting, and Discord reconnects. The project is now at 339,000 GitHub stars.
**Cloudflare RSAC 2026 — MCP Server Portal and AI Security for Apps Now GA** — Cloudflare used the RSA Conference this week to ship two meaningful MCP and AI security products. First: AI Security for Apps reached general availability on March 11 and is getting attention now. It sits as a reverse proxy in front of any AI-powered application, automatically detecting LLM-powered endpoints by behavior rather than URL patterns — which matters because most production AI apps don't follow predictable naming conventions. Endpoint discovery is now free across all Cloudflare plan tiers. Second: Cloudflare One is positioning itself as the first SASE platform with native MCP server security. The MCP Server Portal centralizes request logs, enforces identity-based authentication for every user and agent connection, and aggregates all MCP servers behind a single unified endpoint. With the MCP ecosystem's security posture still poor — recent scans show only two and a half percent of servers pass basic safety checks — having a SASE-layer solution for MCP governance could matter for enterprise deployments where tool sprawl is already a problem.
**Google Nearing $5 Billion Deal to Finance Anthropic's Texas Data Center** — The Financial Times reports Google is close to finalizing a multibillion-dollar financing agreement for a massive data center campus in Texas leased to Anthropic. The facility is being built by Nexus Data Centers on a 2,800-acre site and is targeting 500 megawatts of capacity by early 2026, with a long-term expansion potential of 7.7 gigawatts. Google's role would involve construction loans and facilitating bank lending. This is distinct from Google's equity stake in Anthropic and from the $30 billion Series G raised in February — it's a separate infrastructure financing deal that would deepen the Google-Anthropic relationship at the compute layer. Coming just two days after Judge Lin granted Anthropic a preliminary injunction against the Pentagon's supply-chain blacklist, this signals that Anthropic's enterprise and infrastructure momentum is continuing despite the DoD standoff.
That's the briefing for Sunday, March 29th. ClawCon Tokyo kicks off this week — worth watching for any major OpenClaw ecosystem announcements from the community.