# Agent Platform Research Briefing — February 24, 2026
_Generated: 6:00 AM PT_
---
## ✅ Genuinely New Stories
### 1. Anthropic Accuses Chinese AI Labs of Industrial-Scale Claude Distillation
**Category:** Anthropic
**Sources:** TechCrunch, VentureBeat, The Hacker News, CNN (all published Feb 23–24, 2026)
Anthropic dropped a major bombshell yesterday, accusing three Chinese AI companies — **DeepSeek, Moonshot AI, and MiniMax** — of orchestrating coordinated campaigns to steal Claude's capabilities. According to the filings:
- **24,000 fake accounts** created to circumvent Anthropic's ToS
- **16 million Claude queries** used to extract training data via knowledge distillation
- Anthropic has filed a legal complaint and is calling for stronger U.S. export controls
Distillation itself is common in AI (labs do it to create cheaper versions of their own models), but using fraudulent accounts to distill a competitor's proprietary model is explicitly prohibited. Claude is not available in China, so the accounts used VPN proxies to bypass geographic blocks.
This story lands in the middle of the U.S. AI chip export debate — Anthropic is explicitly citing this as evidence for why tighter controls are needed.
**Note:** Moonshot AI is the same company behind Kimi Claw (previously covered). Their simultaneous alleged distillation campaign adds a new dimension to that story.
---
### 2. MCP SDK Itself Now Has Critical CVEs — Upgrade to v1.26.0
**Category:** MCP Security
**Source:** DEV Community / kai_security_ai (published Feb 24, 2026)
Beyond individual MCP server misconfigurations, the **official @modelcontextprotocol/sdk** TypeScript library has two newly disclosed critical vulnerabilities:
- **CVE-2026-0621** (ReDoS, CVSS High): The `UriTemplate` class contains a regex vulnerable to catastrophic backtracking. An attacker sends a crafted resource URI, hangs the server process before any auth check. Affects all SDK versions ≤1.25.1.
- **CVE-2026-25536** (Cross-Client Data Leak, CVSS High): When a `StreamableHTTPServerTransport` handles multiple clients in stateless deployments, JSON-RPC message ID collisions cause one client's tool execution response to route to a different client's HTTP connection. A second issue with `McpServer` instance reuse silently overwrites the internal transport reference, leaking progress notifications to the wrong client.
**Fix:** Upgrade to `@modelcontextprotocol/sdk v1.26.0`.
This is different from previously covered MCP-uvx pre-execution issues — this is the SDK foundation itself. If you run any TypeScript MCP servers, update now.
---
### 3. Temporal Raises $300M at $5B Valuation — Agent Orchestration Infrastructure
**Category:** Agent Frameworks
**Source:** Reuters, GeekWire (Feb 17, now getting analyst coverage Feb 23–24, 2026)
Temporal $300M Series D** led by Andreessen Horowitz, valuing it at **$5 billion**. The pitch: as AI agents become long-running, stateful processes requiring fault-tolerant execution, Temporal's platform (originally built for cloud workflow orchestration) is becoming essential agent infrastructure.
Already used by Netflix, Stripe, and Coinbase for workflow orchestration. AI agent teams at multiple companies are now building on it for agent reliability — retry logic, state persistence, audit trails.
This is notable as dedicated agent framework investment vs. the all-in-one agent platform plays (OpenClaw, Azure AI Foundry). Temporal is betting on being the plumbing beneath any framework.
---
### 4. Genviral Releases Social Media Automation Skill for OpenClaw
**Category:** OpenClaw Ecosystem
**Source:** Yahoo Finance / press release (Feb 23, 2026)
Genviral six platforms**. Users can instruct their OpenClaw agent to draft, schedule, and post content with platform-specific formatting.
Lower-tier news but worth noting as the OpenClaw skill ecosystem continues to expand commercially.
---
## 📋 No New Developments (Briefly Noted)
| Topic | Status |
|-------|--------|
| OpenClaw 2026.2.19 / CVE patch status | No new patches announced; GLaDOS still on older version |
| OpenClaw Endor Labs CVEs (Feb 23) | No remediation patch announced yet |
| PersonaPlex-7B | No new benchmarks or adoption news |
| MiniMax Speech 2.6 | No new developments |
| MCP/uvx pre-execution injection | No patch released |
| Microsoft Agent Framework RC | Still RC, no GA date |
| Claude Code Security preview | No expansion announcements |
| LangChain observability | No new adoption news |
| OpenAI Realtime API | No new capabilities |
| Kimi Claw | No major news (see Moonshot distillation story above) |
| Claude Opus 4.6 | No new API access news |
| LangGraph / CrewAI / DSPy / AutoGen | No releases in last 24h |
---
## 🎙️ Audio Script (for TTS generation)
See companion MP3 file.