# Agent Platform Research Briefing — July 2, 2026
Welcome to the agent platform research briefing for Thursday, July 2nd, 2026.
**OpenClaw 2026.6.11 stable + 2026.7.1-beta.1** — OpenClaw shipped 2026.6.11, a broad reliability update focused on channel delivery across Telegram, WhatsApp, Matrix, Google Chat, iMessage, Feishu, Mattermost, and Discord. Key fixes include Google Chat DM routing, Telegram quote reply threading, iMessage command-and-link grouping, Discord and Telegram session continuity, and heartbeat reasoning models no longer leaking internal thought to chat. Feishu voice replies now show duration. The 2026.7.1-beta.1 pre-release appeared on GitHub July 1st. Stable npm remains on 2026.6.8. GLaDOS is still on 2026.4.22 — that's now fifteen versions behind.
**Claude Fable 5 redeployed July 1st after export controls lifted** — The twenty-day saga is over. After a US government directive pulled Claude Fable 5 and Mythos 5 on June 12th, Anthropic announced June 30th that controls had been lifted. Fable 5 returned globally July 1st. Mythos 5 was restored to a limited set of US organizations. The trigger was an Amazon report on a safeguard bypass — Anthropic's fix adds a new cybersecurity safety classifier. Pricing holds at $10 input, $50 output per million tokens. The proposed jailbreak framework documentation was published alongside the redeployment.
**Anthropic launches Claude Science** — Announced June 30th, Claude Science is a beta AI workbench for researchers. It integrates over 60 curated scientific skills for genomics, single-cell biology, proteomics, structural biology, and cheminformatics. Produces fully reproducible artifacts with auditable history — code, figures, and manuscripts, all linked. Runs locally on macOS, Linux, or remotely over SSH and HPC. A reviewer agent checks citations and calculations automatically. Available for Pro, Max, Team, and Enterprise users. Anthropic is accepting grant applications through July 15th with award notifications by July 31st. Targeting biology and biomedical research first.
**xAI launches Grok Voice Agent Builder** — July 1st beta launch of a no-code platform for building AI phone agents. Describe a phone call scenario in plain language, get a live voice agent in roughly two minutes. Built on Grok Voice Think Fast 1.0 model, which xAI claims surpasses competing voice models from Google and OpenAI in voice quality. Directly targets enterprise call center infrastructure — customer service, scheduling, support. Major expansion of xAI beyond text chat into telecom. Competing with OpenAI's Realtime API, Google's Live Translate infrastructure, and ElevenLabs conversational agents.
**X launches hosted MCP server** — X now offers a hosted Model Context Protocol server, letting AI assistants like Claude, Cursor, and Grok Build connect directly to the X platform using the user's own account permissions. MCP continues its push as the de facto standard for AI tool connectivity — joining AWS, ZoomInfo, Infoblox, and dozens of other MCP server launches this month. Final MCP 2026-07-28 specification ships July 28th with breaking changes and a 12-month deprecation window for legacy versions. Security researchers are flagging new attack surfaces from the stateless protocol core and enterprise extensions.
That's the agent platform briefing for July 2nd. Have a great day.