# Agent Platforms Research — 2026-03-01
## Summary
Three genuinely new stories today. No major framework releases. MCP security picture deepening with new empirical data. Anthropic Pentagon saga now generating unexpected public goodwill.
---
## NEW STORY 1: Claude Surges to #1 on US App Store — Pentagon Backlash Effect
**Source:** CNBC (Feb 28), TechCrunch (Feb 28), Business Insider (Feb 28)
**Status:** NEW — not previously covered
In a striking public response to the Pentagon standoff saga, Claude surpassed ChatGPT as the #1 free app in the US Apple App Store on Saturday February 28. Users documented switching from ChatGPT to Claude on social media in solidarity with Anthropic's refusal to drop safety guardrails. TechCrunch placed it at #2 (behind ChatGPT) while CNBC and Business Insider confirmed a #1 peak in the overall free apps chart. Ranking fluctuated through the evening. Either way, a dramatic consumer response to a corporate ethics story — and a rare moment where "doing the right thing" produced a measurable commercial benefit for Anthropic.
**Relevance:** Anthropic's safety-first brand is becoming a genuine market differentiator, not just a liability.
---
## NEW STORY 2: MCP Security Audit — 30% of 706 Servers Have No Authentication
**Source:** dev.to/@kai_beget (March 1, 2026, posted ~9 hours ago)
**Status:** NEW angle — SDK CVEs covered (Feb 24), this is deployment-level empirical data
An automated security scanner analyzed 706 publicly reachable MCP servers:
- 30% had zero authentication — tools accessible to anyone on the internet
- 47% had at least one high-severity issue
- Common issues: auth bypass, prompt injection vectors, data exfiltration via error messages, no rate limiting (45%), dangerous tools (delete/send/modify) with no confirmation step
This is distinct from the MCP SDK CVEs (CVE-2026-0621, CVE-2026-25536) covered last week — those were code-level; this is a deployment hygiene survey showing the ecosystem at large is operating without basic security controls. Given MCP servers now connect to databases, file systems, email APIs, and financial services, the exposure surface is significant.
Note: The scanner author identifies as "an autonomous AI security researcher running 24/7" — interesting meta-context.
---
## NEW STORY 3: Firefly Alpha "Stairway to Seven" — Window Opens Today
**Source:** NASASpaceFlight.com, KSBY, Noozhawk (Feb 28 - March 1)
**Status:** UPDATE to firefly-alpha-return-to-flight (first covered Feb 28)
After a Saturday weather scrub, Firefly Alpha FLTA007 "Stairway to Seven" is targeting today, Sunday March 1, with a window from 4:50 PM to 6:50 PM PT from Vandenberg SLC-2. This is Firefly's first Alpha launch in approximately 10 months following an April 2025 anomaly, and the last flight in the current Block 1 configuration before moving to Block 2. Primary goal: nominal first and second stage separation. No launch outcome available at research time (6 AM PT).
---
## Topics Checked — No New Developments
- OpenClaw releases: Latest confirmed is 2026.2.26/2026.2.27 — no new version in last 24h
- LangChain/LangGraph/CrewAI/AutoGen: No new releases or major announcements in 24h window
- Voice AI/Realtime APIs: No new model releases; general market commentary only
- DeepSeek V4: Still not released; do not resurface until release
- GPT-5.3 Garlic: Polymarket still targeting March window; no release
- Starship Flight 12: Ship 39 cryo testing ongoing; no new launch date confirmed