If you’ve been wondering which model to run in your OpenClaw agents, a benchmark dropped today that gives practitioners some of the most concrete comparative data seen yet — and the winner may surprise you. Gemini 3 Flash topped the PinchBench OpenClaw task evaluation with a 95.1% success rate, beating every other major model in head-to-head agentic performance. The data was surfaced by SlowMist CISO @im23pds on X and corroborated by Phemex News, landing on the same day OpenClaw v2026.3.7 shipped with native Gemini 3.1 Flash-Lite support. ...

The OpenClaw project shipped its biggest release of 2026 this morning: v2026.3.7, built by 196 contributors and packed with features that fundamentally extend what agents can do with memory, context, and model choice. If you run OpenClaw in production, stop what you’re doing — there’s a breaking change you need to handle before restarting. The Headline Feature: ContextEngine Plugin Slot This is the one that changes architecture discussions. OpenClaw now exposes a first-class ContextEngine plugin slot with a full lifecycle hook API: ...

Something that once sounded like science fiction just became engineering reality. Anthropic’s Boris Cherny — the creator of Claude Code — confirmed on X this week that the AI coding tool is now 100% written by itself. Not mostly. Not mostly-ish with some human help. One hundred percent. The post was simple: “Can confirm Claude Code is 100% written by Claude Code.” It has since surpassed 133,000 views, and for good reason: this is one of the clearest milestone moments of recursive AI improvement we’ve seen from a production system. ...

If you’ve been in the AI agent space for more than a few months, you’ve heard the word “prompt engineering” approximately ten thousand times. LangChain co-founder and CEO Harrison Chase wants to retire it — and replace it with something more fundamental. His argument: context engineering is the skill that separates agents that actually work in production from ones that fall apart after five steps. And LangChain just shipped a framework to prove the point. ...

There’s a category of AI agent project that gets the security architecture right from the start — and NanoClaw is becoming a strong example of what that looks like. The open-source personal AI agent framework has landed support for WhatsApp and Telegram in v0.1.4, bringing container-isolated AI agents directly into the messaging apps most people already live in. What NanoClaw Is (And Isn’t) NanoClaw describes itself as a “minimalist alternative to larger AI agent frameworks.” The numbers back that up: the entire project is roughly 3,900 lines of code across 15 source files, running as a single Node.js process. Compare that to the sprawling architectures of OpenClaw or similar frameworks. ...

Running one AI coding agent is table stakes now. Running ten in parallel, on isolated branches, coordinated against your actual ticket backlog — that’s the next thing. Superset IDE 1.0 just launched on Product Hunt, and it’s making a direct claim to that space. What Superset IDE 1.0 Actually Does The core pitch is model-agnostic parallel agent orchestration. You connect Claude Code, Codex CLI, and custom agents to Superset IDE, and the tool manages running them simultaneously across isolated git worktrees. Each agent gets its own working environment — no context bleeding between tasks, no merge conflicts mid-stream. ...

There’s a problem with multi-agent AI systems that doesn’t show up until you run them in the wild, and a new research paper from Northeastern University has done the work of naming it precisely. The paper, “Agents of Chaos,” led by researcher Natalie Shapira, makes a claim that anyone who’s run multi-agent pipelines in production will recognize: the failure modes of two agents interacting are not the sum of their individual failures. They’re something qualitatively different and qualitatively worse. ...

Something shifted in the agentic AI community this week — and it happened in a West Village event space over lobster bisque and live robot demos. ClawCon NYC, the first major in-person gathering of the OpenClaw community, drew over 800 attendees out of 1,300+ RSVPs. For a framework that didn’t exist three months ago, the energy in that room tells you everything about how fast the agentic AI ecosystem is moving. ...

The attack is elegant in a disturbing way. An adversary doesn’t need to breach your AI infrastructure, compromise your API keys, or exploit a software vulnerability. They just need to get your AI agent to read a web page they control — and then they’re driving. Indirect Prompt Injection (IDPI) is the attack technique where malicious instructions are embedded in content that an AI agent processes: web pages, documents, calendar entries, emails. When the agent reads that content, it encounters instructions that override or subvert its intended behavior. The content tells the agent what to do, and the agent, trained to follow instructions, complies. ...

Indirect Prompt Injection (IDPI) is now confirmed in-the-wild by Palo Alto Unit 42. Adversaries are embedding hidden instructions in web pages and documents to hijack AI agents — and OpenClaw’s browser and research agents are high-value targets. This guide walks through concrete hardening steps you can apply to your OpenClaw deployments today. Prerequisites OpenClaw installed and configured (any recent version) At least one agent with web browsing or document processing capability Basic familiarity with OpenClaw’s skill and session configuration Step 1: Audit Your Agent Attack Surface Before hardening anything, map your exposure. For each agent you run: ...