Posts

Three critical security vulnerabilities in MCP database integrations have been publicly disclosed — and the fact that one vendor has refused to issue a patch makes this an active, ongoing risk for anyone running AI agents with database access. The findings, reported by The Register, come from a security researcher who spent several weeks systematically examining how popular database MCP connectors handle authentication, authorization, and input sanitization. What they found should prompt an immediate audit for any team using AI agents with database integrations. ...

AI cyber capabilities are accelerating faster than experts predicted — and a new milestone from Anthropic’s Claude Mythos Preview makes that alarmingly clear. The UK’s AI Safety Institute (AISI) has published an evaluation showing that Claude Mythos Preview became the first AI model in history to complete “The Last Ones” (TLO) — a grueling 32-stage corporate cyberattack simulation designed to test frontier AI offensive cyber capability. The results have forced AISI to dramatically revise its timeline projections for AI cyber capabilities. ...

Physical AI just crossed into industrial scale. FANUC — the Japanese robotics giant with the largest installed base of industrial robots on earth — has announced a strategic partnership with Google to integrate Gemini Enterprise into its global fleet of 1.1 million robots. Markets noticed immediately. FANUC shares surged 16% to a record high — approximately 8,880 yen — on the announcement, reflecting investor conviction that this partnership represents a fundamental shift in what industrial robots can do. ...

Android just crossed a threshold that smartphone makers have been inching toward for years: your phone doesn’t just answer questions anymore — it acts on them. At The Android Show: I/O Edition 2026 on May 12, Google unveiled Gemini Intelligence — a comprehensive suite of agentic features that fundamentally reshape how Android operates. This isn’t an AI assistant upgrade. It’s the beginning of Android as an autonomous agent operating system. ...

Microsoft has quietly been running something remarkable inside its security organization: a system called MDASH that orchestrates more than 100 specialized AI agents to autonomously hunt for vulnerabilities in Windows and Microsoft’s broader product surface. This month, MDASH earned its most concrete validation yet — discovering 16 previously unknown Windows vulnerabilities, including 4 Critical remote code execution flaws, all patched in the May 2026 Patch Tuesday update. What Is MDASH? MDASH — Microsoft’s Multi-agent Dynamic Autonomous Security Hunter — isn’t a single AI model. It’s an orchestration system that coordinates a fleet of specialized agents, each optimized for specific security tasks: static analysis, dynamic testing, fuzzing, exploit development, and impact verification. ...

Here’s a number that should stop every AI optimist in their tracks: 74% of enterprises have rolled back live AI customer communications agents after deploying them in production. That’s the headline finding from Sinch AB’s new global research report, “The AI Production Paradox” — and it cuts against the prevailing narrative that enterprise AI is steadily marching toward broad deployment. The Gap Between Experiment and Production Sinch’s research captures a phenomenon that many practitioners have sensed but rarely seen quantified this sharply. There’s a persistent, stubborn gap between what AI agents can do in controlled demonstrations or sandbox environments and what they reliably deliver in live, production customer-facing scenarios. ...

Anthropic is going after the SMB market in a big way. The company announced Claude for Small Business today — a bundled package of connectors, agentic workflows, and reusable skills that brings Claude deeply into the tools small and medium-sized businesses already use. The positioning is direct: this is designed to compete with Microsoft Copilot for Business. What Claude for Small Business Includes The product centers on three core capabilities: ...

After weeks of controversy, Anthropic has reversed course on one of its most contentious policy decisions of the year: the prohibition on using Claude subscriptions to power third-party agents like OpenClaw. The reversal isn’t a clean return to the old model, though. Anthropic has introduced a new credit architecture that separates interactive usage from programmatic agent usage — and that distinction matters a lot for the agentic AI community. What Changed Earlier today, Anthropic’s official developer communications account @ClaudeDevs announced that all paid Claude subscribers now receive a dedicated “Agent SDK” credit pool. This pool is specifically allocated for programmatic uses — including external, third-party agents like OpenClaw and claude -p command-line usage. ...

When the person who built a tool tells you how they use it, that’s signal worth paying attention to. Boris Cherny, the Anthropic engineer who created Claude Code, recently revealed that he runs “a few thousand” AI coding agents overnight — and manages them from his phone. That’s not a hypothetical scale. That’s a practitioner using their own tool at the edge of what it can do. The Overnight Agent Workflow In a Business Insider interview, Cherny described his methodology: rather than running single coding tasks sequentially during his workday, he queues up thousands of parallel agent tasks to run overnight. He wakes up to completed work that would have taken days or weeks to do sequentially. ...

Notion just made a very large bet on the agentic future of productivity software. The company launched a full developer platform today that turns its workspace into an orchestration layer for AI agents — and the feature set goes well beyond a simple API. CEO Ivan Zhao summarized the vision in four words: “Any data, any tool, any agent.” That’s an ambitious positioning statement, and the platform announced today backs it up. ...