OpenClaw v2026.6.9/v2026.6.10: Agent Recovery, Model Routing Overhaul, and Platform Stability
OpenClaw v2026.6.9/v2026.6.10 ships agent recovery, smarter model routing, auto fast mode for chats, and major platform stability improvements.
News
Copilot Studio Agents Now Reason Over Azure Databricks via Single MCP Connection
Microsoft Copilot Studio now connects to Azure Databricks workspaces via a single MCP connection in Public Preview, enabling enterprise data reasoning.
Confluent Report: 32% of Organizations Have Agentic AI in Production — But 77% Report Stalled Projects
Confluent's 2026 report: only 32% of orgs have agentic AI in production, and 77% of those are stalled — data quality and infrastructure are the real bottlenecks.
Anthropic Doubles Down on Enterprise Claude: Cron Scheduling, Credential Vaults, and Okta Zero-Touch MCP Auth Now in Beta
Anthropic adds cron scheduling, credential vaults, and Okta zero-touch MCP auth to Claude Managed Agents — enterprises can now run agents on autopilot securely.
MCP 2026-07-28 Release Candidate: Stateless Protocol, Formalized Extensions, and Authorization Hardening
MCP's biggest spec update since launch drops session IDs for stateless routing, formalizes Extensions, and hardens OAuth — final spec due July 28.
OpenClaw 2026.6.9 Stable Released — Richer Telegram Delivery, Stronger Codex Integration, and Improved Agent Recovery
OpenClaw 2026.6.9 stable: 422 PRs, richer Telegram delivery, stronger Codex integration, improved agent recovery, and new Zalo channel.
Anthropic's Always-On 'Conway' Agent: Webhooks, GitHub Triggers, and the Autonomous Background Assistant That Leaked From Their Codebase
Anthropic's unreleased 'Conway' always-on agent resurfaces: persistent webhooks, GitHub triggers, and autonomous background operation revealed via code leak.
AutoJack: A Single Malicious Webpage Can Execute Arbitrary Code on Developer Machines Running AutoGen Studio
Microsoft discloses AutoJack: a zero-click RCE exploit chain letting a malicious webpage hijack developer machines via AutoGen Studio's MCP WebSocket surface.
MCP Is Now an Active Attack Surface: STDIO Injection, Tool Poisoning, and Rug Pull Attacks in Production
MCP is under active exploitation: STDIO injection, tool poisoning via hidden instructions, and rug pull attacks are documented across Claude Code, Cursor, and VS Code.
OpenClaw v2026.6.9 Released: CI Gating, Fail-Closed Approvals, and Harness Lifecycle Parity
OpenClaw v2026.6.9 ships CI gating and fail-closed approvals: unanswered approval requests now default to 'no'. Here's what changed and what to watch for.
Claude Corps Fellowship Opens 1,000 AI Jobs at Nonprofits Before July 17 Deadline
Anthropic launches $150M Claude Corps: 1,000 paid AI fellowships at nonprofits — no degree required. Applications open now.
Microsoft AutoJack: A Single Malicious Webpage Can RCE the Host Running Your AI Agent
Microsoft's AutoJack exploit shows a malicious webpage can hijack a browsing AI agent to execute code on the host via local MCP endpoints — a systemic architecture risk.
Claude Code Artifacts Beta: AI Coding Sessions Become Live, Shareable Browser Pages
Anthropic's Claude Code Artifacts beta lets enterprise teams publish AI coding sessions as live, shareable HTML pages with real-time updates and version history.
Fable 5 System Prompt Leaked on GitHub — 120,000 Characters of Hidden Rules Exposed
A 120,000-character system prompt purportedly from Claude Fable 5 circulates on GitHub, though Anthropic contests its authenticity — reigniting AI transparency debates.
OpenClaw v2026.6.8 Released — Telegram Rich Messaging, GLM-5.2 + Haiku 4.5, Parallel Search Opt-In
OpenClaw 2026.6.8 ships Telegram rich text, GLM-5.2 and Haiku 4.5 model support, and security-tightened search opt-in across 192 changes.
The Information: Hermes Agent (Nous Research) Eclipses OpenClaw by Some Measures
The Information reports Nous Research's Hermes Agent surpassed OpenClaw in daily OpenRouter token volume — a direct competitive signal for the personal AI agent space.
JPMorgan Cuts Off Claude (Anthropic) Access for Hong Kong Staff — Follows Goldman Sachs
JPMorgan blocks Hong Kong staff from Claude AI access — following Goldman Sachs in April — as US-China tensions reshape enterprise AI adoption across Wall Street.
CrewAI v1.14.8 Alpha — FlowDefinition DSL with if/each/DMN, Human Feedback, JSON-First Crews
CrewAI v1.14.8 alpha drops FlowDefinition DSL upgrades including DMN business rules, if/each expressions, and declarative human feedback loops.
Hermes Agent v1.0+ Async Subagents — delegate_task_async Enables Non-Blocking Background Agent Delegation
Hermes Agent (Nous Research) adds async subagents via delegate_task_async, enabling non-blocking long-running background task delegation.
OpenClaw v2026.6.8 Released — Richer Telegram/WhatsApp, New Models, Security Defaults Hardened
OpenClaw v2026.6.8 lands with richer Telegram/WhatsApp formatting, GLM-5.2 and Claude Haiku 4.5 support, and keyless search now opt-in.
Swarms 'Bundles' — Shareable Multi-Agent Workflow Packages Combine Marketplace Agents + Custom Prompts
Swarms Bundles lets users package multi-agent workflows with custom prompts into shareable single-link toolkits — like npm for agent systems.
Fast Company: How Peter Steinberger Built OpenClaw
Fast Company's AI 20 for 2026 spotlights Peter Steinberger and how OpenClaw reached 500,000 global deployments.
Low-Skilled Attacker Used Claude and Codex to Breach 14 Companies
A low-skilled attacker used Claude Code and Codex to breach 14 companies — the first documented commodity AI agent cyberattack.
OpenClaw 2026.6.8 Released — 160+ PRs, Smarter Telegram/WhatsApp, Major Stability Improvements
OpenClaw 2026.6.8 drops with 160+ PRs: richer Telegram messages, WhatsApp fixes, new model support, iOS reconnects, and more.
OrcaRouter AI Threat Report 2026: Agents Are Getting Socially Engineered — Free Firewall Launched
OrcaRouter's AI Threat Report 2026 reveals agents are being socially engineered — and ships a free open-source firewall to fight back.