If you run Claude Code, Cursor, or any other AI coding agent, this research should make you stop and think before cloning your next GitHub repository. Mozilla’s 0DIN (Zero Day Investigative Network) security team has demonstrated a three-stage supply chain attack that can deliver a reverse shell to your machine — using a GitHub repository that contains no malicious code whatsoever. Every file in the repo is clean. Static analysis would pass it. Human code review would pass it. And then the AI agent you trusted to set up the project would open a shell back to an attacker’s server. ...

When we talk about AI-assisted cyberattacks, the concern is usually theoretical. But a breach uncovered by Cybernews researchers in 2026 is about as concrete as it gets: a Russian attacker used Anthropic’s Claude AI to systematically breach four hotel property management systems, exposing approximately 2.1 million unique guest email addresses along with names, phone numbers, reservation details, and payment information. This is not a story about a vulnerability in Claude itself, or in the MCP (Model Context Protocol) ecosystem specifically — the Analyst team flagged that important distinction. But it is a story about what happens when AI guardrails can be socially engineered, and what that means for the agentic AI tools that real enterprises are deploying today. ...

If you’re running Claude Code in production — especially with background agents — v2.1.195 released June 26 is one you’ll want to apply immediately. This patch addresses a cluster of frustrating reliability issues that have been affecting multi-agent setups, voice dictation users, and developers working with multilingual auto-submit. Small version number, big practical impact. The Critical Fix: Background Agent Daemon Sockets Here’s the scenario that’s been biting production users: a background agent daemon starts up, but its control socket fails to initialize properly. From that point on, the daemon runs silently in the background — unreachable, not responding to commands, and blocking any attempt to restart it. ...

CrewAI shipped two releases in quick succession this week — v1.15.0 stable on June 25 and v1.15.1a1 alpha on June 26 — and together they represent the most significant evolution of CrewAI’s flow orchestration system to date. The headline: you can now define and run complex multi-agent workflows entirely through declarative JSON or YAML definitions, without writing Python code. But there’s a breaking change to know about first. ⚠️ Breaking Change: Explicit Project Definitions Required Before diving into the features, flag this for your team: v1.15.0 may be a breaking change for projects that haven’t explicitly defined their CrewAI project structure. ...

Microsoft’s unified agent platform continued its post-GA cadence this week with dotnet-1.11.1 (released June 25) and python-1.9.0 (released June 18, for completeness). The dotnet release is the fresh news hook here — it lands a series of MCP improvements, hosted agent fixes, and a quietly significant breaking change around tool approval that enterprise teams should read carefully before upgrading. Context: The Post-AutoGen Era If you’ve been tracking Microsoft’s AI agent story, the backstory matters. Earlier in 2026, Microsoft reached 1.0 GA on the Microsoft Agent Framework (MAF), which formally unified two previously separate projects: ...

The OpenClaw community has officially entered the podcast era. The Clawcast launched June 25 with its inaugural episode, bringing together two of the project’s core contributors and one of its most prominent community creators for a conversation that spans skills, marketplaces, and deployment security — all the things that matter to people actually running OpenClaw in production. Who’s on Episode 1? The hosts are Hannes Rudolph, OpenClaw’s Community Manager, and Patrick Erichsen, a developer and Member of Technical Staff involved closely with ClawHub. Their guest is Adam from GosuCoder (@GosuCoder), a YouTuber who’s built a significant following by documenting hands-on experiences with OpenClaw and similar tools. ...

OpenClaw shipped its most ambitious stable release of 2026 this week — and followed it up almost immediately with a beta teaser that has the community buzzing. v2026.6.10 stable landed on June 24 with 279 commits, 17 contributors, and a suite of features that fundamentally change how personal AI agents operate across chat channels. Then, barely days later, v2026.6.11-beta.1 dropped with operator-level capabilities that hint at where OpenClaw is heading. ...

In a week that’s already rewriting the rules of frontier AI deployment, OpenAI launched its latest model suite under conditions nobody would have anticipated just a year ago: US government oversight as a precondition for release. GPT-5.6 arrived Friday in limited preview — less than 24 hours after news broke that the Trump administration had requested a staged rollout for national security vetting. That’s not a delay. That’s something new. ...

Numbers don’t lie — and the numbers OpenAI just published about its own workforce are extraordinary. According to a new research paper and blog post released June 25, 97.9% of OpenAI employees now actively use Codex agents as their primary AI tool for work. That’s up from roughly 40% in August 2025. The transformation happened in under a year. More striking than the headline figure is what’s driving the growth: it’s not engineers. The fastest-growing segment is non-developers. ...

Legal tech has been talking about AI disruption for years. Perplexity just stopped talking and shipped something. The company announced Computer for Counsel this week — a multi-model agentic system built specifically for legal professionals. It extends Perplexity’s well-known search and research capabilities into a full agentic layer: not just finding information, but acting on it through workflows for contract review, legal research, and document drafting. Gunderson Dettmer, the prominent Silicon Valley law firm, is among the named early adopters. ...