# Agent Platform Research Briefing — March 8, 2026
**Research window:** Last 24–48 hours
**Generated:** 2026-03-08 06:00 AM PT
---
## 1. OpenClaw 2026.3.7 Released — Context Engine Plugin Architecture + Persistent ACP Bindings
OpenClaw shipped version 2026.3.7 on March 8, 2026 at 05:52 UTC. Founder Peter Steinberger announced the beta build on X earlier the same day as "Beta 3.7."
Key changes in this release:
- **Context Engine plugin interface**: New `ContextEngine` plugin slot with full lifecycle hooks — `bootstrap`, `ingest`, `assemble`, `compact`, `afterTurn`, `prepareSubagentSpawn`, `onSubagentEnded`. A slot-based registry enables config-driven resolution. This lets plugins like `lossless-claw` provide alternative context management strategies without modifying core compaction logic. Zero behavior change when unconfigured.
- **ACP/persistent channel bindings**: Durable Discord channel and Telegram topic binding storage survives restarts. Per-topic `agentId` overrides in forum groups and DM topics allow dedicated agents with isolated sessions per topic.
- **Telegram ACP spawn improvements**: Accept Unicode dash option prefixes in `/acp spawn`, Telegram topic thread binding (`--thread here|auto`), actionable approval buttons with prefixed approval-id resolution, pin successful bind confirmations in-topic.
- **TTS/OpenAI-compatible endpoints**: New `messages.tts.openai.baseUrl` config with endpoint-aware directive validation — allows routing TTS to custom-hosted OpenAI-compatible TTS endpoints.
- **Compaction lifecycle events**: New `session:compact:before` and `session:compact:after` internal events plus plugin compaction callbacks, so automations can react to compaction runs.
- **Configurable post-compaction sections**: `agents.defaults.compaction.postCompactionSections` lets deployments choose which `AGENTS.md` sections re-inject after compaction.
- **Perplexity web search**: Switched to Search API with structured results and new language/region/time filters.
- **iOS App Store Connect prep**: Bundle identifiers aligned under `ai.openclaw.client`, Watch app icon refresh, Fastlane metadata/screenshot automation.
- **Docker/Podman**: New `OPENCLAW_EXTENSIONS` env for preinstalling extension npm dependencies at image build time.
**Significance:** The Context Engine plugin slot is the largest architectural addition since ACP. It opens the door for pluggable memory backends and advanced context strategies without forking core.
**Sources:** GitHub releases API (github.com/openclaw/openclaw), Odaily (March 8), PANews (March 8), Phemex (March 8)
---
## 2. China OpenClaw Adoption Surge — Shenzhen Government Drafts Support Policy
China's OpenClaw moment is intensifying. On March 8, the AI (Robotics) Bureau of Longgang district in Shenzhen released a draft policy for public comment proposing government support measures for the use of OpenClaw — a signal that the platform is being considered for official institutional backing in China's most AI-active city.
Earlier in the week, Tencent engineers set up free OpenClaw installation booths at Tencent HQ's North Plaza in Shenzhen, drawing large crowds including non-developers. The Indian Express ran a piece headlined "Why 60-year-olds in China are queuing up to learn OpenClaw." Multiple Chinese financial outlets (Futunn, PANews, Jianshi) are framing OpenClaw as the centerpiece of a national AI productivity wave.
**Context:** OpenClaw crossed 250,000 GitHub stars last week. China represents a significant and growing share of the user base, with both Alibaba (Qwen 3.5 integration) and Tencent showing active ecosystem engagement. The Longgang draft policy is the first known government-level action to formally propose support for OpenClaw adoption.
**Sources:** Wikipedia (March 8 edit), Futunn News, PANews, Indian Express, Jianshi
---
## 3. Anthropic Sues Over Pentagon "Supply Chain Risk" Designation
The Anthropic–Pentagon story has a significant new legal chapter. On March 5, the Department of Defense formally notified Anthropic that the company and its products have been designated a "supply chain risk" under federal procurement law — a designation that bars DoD agencies from using Claude systems. Anthropic CEO Dario Amodei confirmed receipt and said the company has "no choice" but to challenge the designation in court, arguing the action exceeds statutory authority under 10 U.S.C. § 3252, which requires the "least restrictive means" to protect the supply chain.
Amazon and Microsoft both issued statements confirming that Anthropic products remain available to their non-DoD cloud customers. Microsoft said its lawyers reviewed the designation and determined it does not apply outside defense contexts.
CNBC separately reported that Claude is still being used by Iranian government entities via third-party API resellers, complicating the DoD's framing of Anthropic as a supply-chain risk on national security grounds.
**Note:** This is a new legal escalation distinct from the previously-covered OpenAI contract swap. The formal court challenge is new as of this briefing window.
**Sources:** CNBC (March 5–8), windowsforum.com, michaelrcronin.com
---
## 4. OpenAI Robotics Chief Resigns Over Pentagon Deal
Caitlin Kalinowski, OpenAI's head of robotics and consumer hardware, resigned on March 7 citing concerns about OpenAI's agreement with the Department of Defense. In a post on X, she criticized the company's speed in partnering with the Pentagon without, in her view, adequately investigating guardrails against mass surveillance and autonomous weapons.
Kalinowski previously led Meta's Project Nazare/Orion AR glasses program. At OpenAI she led both the robotics team and the consumer hardware group (the $200–300 smart speaker and glasses initiatives reported last week).
OpenAI issued a statement: "We believe our agreement with the Pentagon creates a workable path for responsible national security uses of AI while making clear our red lines: no domestic surveillance and no autonomous weapons."
**Significance:** Kalinowski is the highest-profile OpenAI departure directly tied to the military AI debate. Her departure from the robotics/hardware group raises questions about continuity for the consumer hardware roadmap.
**Sources:** Reuters (March 7), TechCrunch, Business Insider, Fortune, Engadget
---
## 5. Updates on Active Stories
**Firefly Alpha "Stairway to Seven":** Launch window updated to March 10 at 12:50 AM PDT from Vandenberg Space Force Base. Previous window (March 9 5:50–7:50 PM PDT) has passed without launch. This is now the fifth attempt for FLTA007.
**Starship Flight 12:** SpaceX confirmed Ship 39 cryo testing completion publicly on March 8. Static fire attempt for Booster 19 / Ship 39 is projected for March 8–10. If static fire succeeds, Flight 12 launch could follow in the second half of March.
---
## Stories Skipped (Nothing New)
- DeepSeek V4: Still unreleased as of March 8 (Integrated Cognition, March 7: "looming entrant rather than a fully documented public launch"). Do not resurface.
- LangChain Deep Agents: Covered March 7. No new development.
- Gemini 3.1 Pro: Covered March 5. Gemini 3 Pro Preview deprecated March 9, but this is a minor migration note, not a new story.
- MCP security: No new CVEs or exploits in the last 24h.