Good morning, Rich. Here's your agent platform research briefing for Wednesday, July 15th, 2026.

**OpenClaw 2026.7.1 ships — ClawStat.us says skip it** — OpenClaw v2026.7.1 went stable this week, bringing major Control UI and onboarding overhauls, updated official iOS, Android, and macOS apps, and expanded model support including GPT-5.6, Tencent Hy3, and Meta Muse Spark 1.1. There's stronger Codex and connected coding-agent integration too. But ClawStat.us rates it "skip this version" with medium confidence after assessment on July 15th. The release has over 60 reported issues including critical gateway crash-loops from legacy memory sidecar migration conflicts, a core build regression leaving 11 workspace packages without build scripts, and gateway restart failures. The previous stable lane, 2026.6.11, remains the safe bet until 2026.7.2 drops.

**Hackers scanning for MCP servers and AI credentials** — A widespread internet scanning campaign is actively probing for exposed Model Context Protocol servers, AI assistant configuration files, and unauthenticated local LLM endpoints. Security researchers reviewed two weeks of Apache and ModSecurity logs from a web host that doesn't even run AI services — and still found roughly 200 requests targeting AI infrastructure. Attackers are sending protocol-aware POST requests to MCP endpoints, probing legacy SSE transports, and using HTTP HEAD requests to fingerprint credential files like dot-env and Claude config files without downloading them. The takeaway: AI infrastructure has graduated to routine reconnaissance targets, right alongside WordPress XML-RPC and Spring Boot Actuator probes.

**Citrix NetScaler adds MCP Gateway** — Citrix announced on July 9th that NetScaler now includes MCP Gateway capabilities, letting enterprises centrally route, authenticate, monitor, and control AI agent requests flowing to backend MCP servers. It's the same playbook that guarded APIs for a decade, now applied to agentic AI traffic. Citrix also expanded NetScaler AI Gateway with policy-based LLM routing and token-level usage visibility. This follows Spring AI and MCP taking center stage at UberConf 2026 this week in Denver, where Java developers discussed governing production AI adoption — not debating whether to adopt it. Spring AI has become the de facto standard for AI in Spring Boot apps, and MCP has crossed 97 million monthly SDK downloads. The pattern is clear: enterprise adoption is racing ahead of governance.

**Anthropic in talks with Samsung for custom inference chips** — Reports emerged that Anthropic is in preliminary discussions with Samsung Foundry to manufacture custom AI inference chips at 2-nanometer process node. Neither company has officially confirmed the deal, but the motivation is concrete: Anthropic pays xAI roughly 1.25 billion dollars per month for Colossus compute, and custom silicon would directly target that spend. Samsung is also reported to be in talks with Google and Meta on similar custom chip deals. If Anthropic follows OpenAI's Jalapeño playbook, this could reshape the economic balance between frontier labs. The Samsung angle is notable because Samsung's foundry business has been losing share to TSMC — winning a marquee AI customer would be a major reversal.

**Starship Flight 13 targeting July 16** — SpaceX is targeting no earlier than July 16th at 6:45 PM Eastern for Starship Flight 13 — the second flight of the Version 3 configuration and the first to deploy Starlink V3 satellites. The Super Heavy booster was rolled back for final checkouts after Wednesday's attempt. Flight 13 will test satellite deployment, upgraded flight systems, and booster recovery. This is the launch the FAA investigation from Flight 12 has been clearing the way for.

That's the briefing for today. Stay sharp.