Posts

Security researchers at Tenet Security have disclosed a new attack class they’re calling “agentjacking” — and if you’re using Claude Code, Cursor, or any AI coding agent that can access your error monitoring system, you need to read this. The core insight is uncomfortable: AI coding agents trust error reports. And Sentry DSNs — the keys that allow your frontend to post error events — are public by design. That combination creates a novel attack surface that requires no malware, no phishing, no credential theft, and no access to your codebase. ...

The Model Context Protocol has published its 2026-07-28 Release Candidate — and if you’re building MCP servers, clients, or skills, you need to understand what’s changing. The lead maintainers, David Soria Parra and Den Delimarsky, describe it as “the largest revision of the protocol since launch.” That’s not marketing language. This RC contains breaking changes and restructures the protocol’s core architecture. The final spec ships July 28. SDK maintainers and skill authors have until then to validate their implementations. With the MCP Dev Summit arriving in Mumbai on June 14-15 (co-hosted by the Linux Foundation’s Agentic AI Foundation), this RC is the live topic of conversation for the MCP ecosystem right now. ...

OpenClaw’s latest pre-release — v2026.6.7-beta.1 — just dropped, and it’s a meaningful update for anyone running agentic pipelines in production. The headline addition is native support for Kimi K2.7 Code, Moonshot AI’s latest open-weight coding model, but there’s plenty more to pay attention to: a Feishu context-leak fix, WebSocket payload hardening, and a string of channel delivery improvements that touch Slack, Telegram, and image attachments. Here’s a breakdown of everything in this pre-release. ...

A new benchmark study from researchers at Nanyang Technological University, ST Engineering, IBM Research, and the University of Illinois Urbana-Champaign has produced a finding the agentic AI community can’t ignore: not a single tested AI web agent consistently blocked prompt injection attacks. Not GPT-5. Not Gemini. Not any configuration tested. The research, conducted via a framework called StakeBench, executed 3,168 adversarial simulations and found that the problem is worse than many practitioners have assumed. ...

Three days. That’s how long Anthropic’s most powerful publicly available AI model remained accessible before the U.S. government stepped in and shut it down. On June 9, 2026, Anthropic launched Claude Fable 5 — a “Mythos-class” model described as exceeding the capabilities of anything Anthropic had ever made generally available. Software engineering, scientific research, vision tasks, complex knowledge work — Fable 5 was state-of-the-art across nearly every benchmark. Alongside it, Claude Mythos 5 launched quietly for vetted U.S. government partners through Project Glasswing, described as carrying “the strongest cybersecurity capabilities of any model in the world.” ...

When some of the most powerful institutions in AI research collectively put $10 million behind a single question, it’s worth paying attention to what that question is. On June 11, 2026, Google DeepMind, Schmidt Sciences, the Cooperative AI Foundation, and the UK’s Advanced Research and Invention Agency (ARIA) announced “Scaling AI Safety for a Multi-Agent World” — a global research funding call targeting the emergent risks that appear when large numbers of AI agents interact at scale. ...

Moonshot AI dropped a serious open-weight coding model on June 12, and the numbers are hard to ignore. Kimi K2.7-Code is a 1-trillion-parameter Mixture-of-Experts model — only 32 billion parameters active per token — purpose-built for agentic coding and long-horizon software engineering. It’s available on HuggingFace, running live on Cloudflare Workers AI, and licensed for commercial use. This is exactly the kind of open-weight release the agentic development community has been waiting for. ...

The question of whether AI agents could ever handle real money just got answered. In June 2026, both Mastercard and Visa launched dedicated payment infrastructure for AI agents — not as a pilot program, not as a proof of concept, but as production services with real partner networks. Mastercard’s Agent Pay for Machines (AP4M) launched on June 10 with 30+ industry partners. Visa’s Intelligent Commerce Connect reached general availability, completing a rollout that began in 2025. Together, they represent an inflection point: the moment agentic commerce became infrastructure. ...

If you’re running Claude Code in CI/CD pipelines, GitHub Actions, headless scripts, or any programmatic mode — you need to read this before June 15, 2026. Anthropic is splitting how Claude Code usage is billed. Starting June 15, programmatic usage via the Agent SDK, claude -p headless mode, GitHub Actions integration, and third-party apps authenticating through the Agent SDK will move to a separate monthly credit pool that does not count against your main Claude plan’s usage limits. ...

Claude Code v2.1.172, released June 10, 2026, is a focused but consequential update that unlocks one of the most requested capabilities for complex agentic workflows: recursive sub-agent nesting up to five levels deep. Alongside that, the release brings improved Amazon Bedrock integration, a plugin marketplace search bar, and critical fixes for 1M-context sessions. Five Levels of Sub-Agent Nesting This is the feature that most Claude Code power users have been waiting for. Prior to v2.1.172, sub-agent spawning was available but depth was limited or unpredictable in practice. Now, Claude Code explicitly supports hierarchical sub-agent trees up to five levels deep. ...