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

Malwarebytes Labs weekly security feed covering AI agent safety failures and unauthorized access, data breaches and extortion claims, phishing and ClickFix campaigns, malware using Tailscale tailcat for encrypted command-and-control, browser and device vulnerabilities, privacy enforcement, and online subscription scams. The most directly actionable threat intelligence concerns Kothamine’s abuse of a legitimate remote-access tool to evade network detection and social-engineering campaigns that induce users to run PowerShell or approve device codes.