Multi-Agent AI Interactions Trigger DoS Cascades, Server Destruction — 'Agents of Chaos' Study
New peer-reviewed study finds multi-agent AI interactions cause server destruction and DoS cascades — most safety evals miss this entirely.
News
Claude Cowork Now Runs Scheduled Recurring Tasks Autonomously While You Sleep
Anthropic's Cowork desktop app now runs recurring scheduled tasks autonomously — a direct answer to OpenClaw-style always-on agent automation.
MiniMax Launches MaxClaw: One-Click Cloud OpenClaw Deployment in 10 Seconds
MiniMax launches MaxClaw — one-click cloud OpenClaw deployment in 10 seconds via the MiniMax Agent platform. No servers, no config required.
OpenClaw v2026.2.26 Released: Cron Reliability, External Secrets, Agent Lifecycle Fixes
OpenClaw v2026.2.26 ships cron reliability fixes, external secrets management, WebSocket Codex, and 11 security hardening patches.
Apple Releases Xcode 26.3 With Autonomous AI Coding Agents (Claude + Codex)
Apple officially ships Xcode 26.3 with native support for Claude and Codex coding agents via MCP — a first for a major OS vendor.
IronCurtain: Open-Source Project Secures and Constrains AI Agents to Prevent Rogue Behavior
IronCurtain is a new open-source wrapper that sandboxes LLM agents, enforces plain-English policies, and prevents rogue behavior — built by Niels Provos.
Microsoft Launches Copilot Tasks: Autonomous Background Agent With Its Own Browser & Cloud PC
Microsoft's Copilot Tasks gives users a cloud PC-powered autonomous agent for scheduled background tasks — rivaling OpenAI Operator.
Microsoft Research Introduces CORPGEN: Multi-Horizon Hierarchical Planning and Memory for AI Agents
Microsoft Research's CORPGEN delivers 3.5x better multi-horizon task completion for AI agents using hierarchical planning and persistent memory.
Oasis Security Discovers Critical Vulnerability Chain in OpenClaw Enabling Full Workstation Compromise
Oasis Security found a critical OpenClaw vulnerability chain enabling full workstation compromise from a browser tab — 40K+ instances exposed.
Anthropic Launches Claude Agent Skills: Modular Reusable Skill Packs for Claude Agents
Anthropic's new Agent Skills framework lets developers build modular, reusable capability packs for Claude — and a third-party marketplace ecosystem is already emerging.
Confluent Adds A2A Protocol Support for Multi-Agent Streaming Data Networks
Confluent adds A2A protocol support to connect AI agents to real-time Kafka streams, marking a major enterprise infrastructure milestone for multi-agent interoperability.
OpenClaw v2026.2.25 Released: Subagent Delivery Refactor, 30+ Security Fixes, Android Streaming Upgrade
OpenClaw v2026.2.25 ships today: 30+ security fixes, subagent delivery state machine refactor, Android streaming upgrade, and a breaking heartbeat policy change.
Anthropic Acquires Computer-Use AI Startup Vercept to Enhance Claude's Agent Capabilities
Anthropic acquires Seattle computer-use startup Vercept, bringing its 9-person team into Claude to accelerate AI agents that operate software.
Critical Claude Code Flaws Allow Remote Code Execution and API Key Theft
Check Point found critical Claude Code CVEs enabling RCE and API key theft via malicious repos — every developer using Claude Code is at risk.
Perplexity Launches 'Computer': A Multi-Agent Super-Agent Orchestrating 19 AI Models
Perplexity launches 'Computer', a super-agent orchestrating 19 AI models to handle end-to-end research, code, and design workflows.
GitHub Engineering Blog: Why Multi-Agent AI Workflows Fail in Production (and How to Fix Them)
GitHub engineers reveal why most multi-agent workflows fail in production — and the three structural patterns that actually fix it.
After OpenClaw Backlash, Quill Bets on Security-by-Design Agentic AI
After OpenClaw's inbox deletion incident, 98% of CISOs are slowing AI agent adoption. Quill pitches security-by-design as the answer.
Anthropic Claude Cowork Launches Enterprise Plugin Marketplace: Finance, HR, Design, Excel, PowerPoint
Anthropic's Claude Cowork gets a full enterprise plugin marketplace with finance, HR, and design connectors for Excel, PowerPoint, and more.
Anthropic Accuses DeepSeek, Moonshot, MiniMax of Industrial-Scale Distillation: 16M Queries, 24K Fake Accounts
Anthropic accuses three Chinese AI firms of using 24K fake accounts to run 16M Claude queries in an industrial-scale model distillation attack.
Google Bans Antigravity Users Routing Gemini Tokens Through OpenClaw OAuth Integration
Google ban-waved Antigravity users routing Gemini tokens through OpenClaw, suspending paid subscribers with no warnings or refunds.
OpenClaw v2026.2.24 Released: Multilingual Stop Phrases, Android Overhaul, Trust Model Hardening
OpenClaw v2026.2.24 ships multilingual emergency stops, a native Android overhaul, and trust model hardening — two breaking changes included.
Meta Director Summer Yue's Inbox 'Speedrun Deleted' by OpenClaw Agent After Compaction Wipes Safety Instruction
Meta's AI alignment director lost 200+ emails when her OpenClaw agent lost its safety instructions during compaction — a guardrails lesson for everyone.
Intuit and Anthropic Partner for Custom AI Agents in Mid-Market Finance — Spring 2026 Rollout
Intuit partners with Anthropic for multi-year deal bringing custom AI agents to QuickBooks and TurboTax, with Claude Code rolling out org-wide.
OpenClaw 2026.2.23 Released: Claude Opus 4.6 Support and SSRF Policy Overhaul
OpenClaw 2026.2.23 ships Claude Opus 4.6 and a breaking SSRF policy shift to trusted-network mode — self-hosted users must migrate.
Twilio Launches A2H: Open Protocol to Standardize Agent-to-Human Workflows
Twilio's A2H is an open protocol for auditable agent-to-human handoffs — and it already has an OpenClaw integration tutorial.