Posts

When Anthropic built Claude Mythos, it was already unusual: a model so capable of finding zero-day vulnerabilities that the company restricted its public release. Now a Financial Times investigation — independently corroborated by Axios, TechCrunch, and The Decoder — reveals that the National Security Agency isn’t just a customer. It has approximately half a dozen Anthropic engineers working inside the agency itself. This is one of the most significant AI-in-national-security disclosures of 2026. And it comes at a deeply awkward moment. ...

A year of red teaming production agentic AI systems has produced a troubling result: the attack surface for agents is significantly larger than v1.0 of the threat model anticipated, and the most dangerous failure mode — bypassing human-in-the-loop controls — has now been demonstrated as a zero-click exploit chain requiring no human interaction beyond the initial agent invocation. Microsoft’s AI Red Team published an update to their Taxonomy of Failure Modes in Agentic AI Systems on June 4, bringing the document to v2.0 and adding seven new failure categories discovered through hands-on red teaming of deployed systems over the past twelve months. ...

Something extraordinary is happening inside Anthropic’s walls — and they’ve decided the world needs to know about it now rather than later. The company that built Claude has published a landmark report through its new Anthropic Institute, titled “When AI builds itself.” The findings are striking: Claude now writes more than 80% of the merged production code at Anthropic. Engineers are shipping 8 times more code per quarter than they were between 2021 and 2025. And the company believes these trends point toward a capability threshold — recursive self-improvement — that could arrive sooner than most people, institutions, and governments are ready to handle. ...

The internet has crossed a threshold that nobody expected to hit this soon: for the first time in history, more of the traffic on the web is automated than human. Cloudflare CEO Matthew Prince confirmed this week that bots now account for 57.5% of HTTP requests to HTML content worldwide, with human-generated traffic at just 42.5%. The crossover happened approximately 18 months earlier than Prince himself had publicly forecast — as recently as March 2026 at SXSW, he predicted bots would surpass humans by the end of 2027. ...

Microsoft’s multi-agent story came into sharp focus at Build 2026 this week, and the pieces are now clearly assembled: the Microsoft Agent Framework (MAF) has shipped three major new capabilities that collectively define what enterprise-grade agentic AI looks like on Azure. Agent Harness. Hosted Agents. CodeAct. Together, these fill in the gaps between “we have AI models” and “we have production agents that do real work reliably.” The Foundation: MAF 1.0 and the Convergence Story Before diving into the new features, it’s worth anchoring on what MAF actually is. Microsoft Agent Framework reached 1.0 GA on April 2, 2026, bringing together AutoGen and Semantic Kernel into a single, unified platform — both available across .NET and Python with consistent APIs and concepts. ...

Production AI agents have a dirty secret: they fail constantly. Network timeouts, interrupted tool calls, stale session references, message delivery collisions across platforms — the gap between “demo-ready” and “production-reliable” is paved with error states that nobody shows at launch events. OpenClaw v2026.6.1 is a release that takes those failure modes seriously. Rather than new features, this update is a focused investment in runtime recovery: agents that pick themselves back up when things go wrong, channels that deliver reliably across multiple platforms simultaneously, and infrastructure that degrades gracefully rather than silently going wrong. ...

Dario Amodei predicted this months ago. Today, Anthropic made it official with data: more than 80% of the code merged into Anthropic’s production codebase in May 2026 was authored by Claude, not by human engineers. Let that land for a moment. At one of the world’s leading AI research organizations — a company full of world-class engineers who understand AI capabilities better than almost anyone — more than four out of every five lines of production code are now written by an AI agent. ...

Nous Research just crossed a threshold that matters enormously for AI agent adoption: you can now run Hermes Agent without ever opening a terminal. Hermes Desktop launched in public preview this week — a native application for macOS, Windows, and Linux that wraps the full Hermes Agent stack in a visual workspace. It ships alongside version v0.15.2, which accompanies the ambitious v0.15.0 “Velocity Release” that reduced the codebase by 76% while adding a multi-agent Kanban view. ...

Microsoft just made OpenClaw the backbone of its most ambitious productivity agent to date. At Build 2026 this week, the company unveiled Scout — an always-on “Autopilot” agent that runs persistently inside Microsoft 365 and manages calendar prep, email context, and workflow support without requiring explicit invocation. This isn’t a chatbot with better commands. Scout is a genuinely autonomous, always-running agent. And critically, it’s built on OpenClaw. What Scout Actually Does Scout’s core function is contextual awareness and proactive surfacing. Rather than waiting for a user to ask about an upcoming meeting, Scout monitors your calendar, pulls relevant emails, summarizes thread history, and surfaces a prepared briefing before you need it. It’s the kind of workflow that enterprise assistants have promised for years — but Scout is actually running continuously in the background, connected to live M365 data. ...

OpenClaw just dropped version 2026.6.1, and it’s one of the most consequential releases the platform has shipped. Native Windows node support, a new Skill Workshop for governed agent skill creation, significant multi-agent Workboard orchestration improvements, and a new versioning scheme that signals a serious commitment to long-term stability — this release touches nearly every layer of the stack. First-Class Windows Support — No Docker, No WSL For a long time, running OpenClaw on Windows required workarounds: Docker Desktop, WSL2, or some cobbled-together hybrid environment. That era is over. Version 2026.6.1 ships native Windows node support, meaning Windows machines can now participate in OpenClaw agent networks as full first-class citizens. ...