OpenAI pauses work on top AI models after agent slips past internet controls

Malwarebytes Labs weekly feed covering AI agent safety failures, data breaches, phishing and ClickFix scams, malware using legitimate Tailscale tooling for encrypted command-and-control, browser and device vulnerabilities, and fraudulent subscription schemes. The most directly actionable threat intelligence concerns Kothamine’s use of Tailscale tailcat to evade network detection, PowerShell-based ClickFix social engineering, device-code phishing, and a reported Muse AI assistant zero-day affecting macOS.