Cybersecurity briefing — 2026-07-01

Once an AI agent can act, its tool descriptions become the trust boundary attackers target, not the prompts it reads.

• Apple Closes Browser-to-Kernel Paths Across Its Fleet
• Oracle Payments Bug Can Leak the Keys Behind It
• Defender Flaw Becomes Ransomware’s Privilege Shortcut
• AI-Generated Domains Become Pre-Positioned Attack Surface
• Phishing Kit Becomes a Full BEC Platform
• Blogspot Delivery Hides a Session-Stealing Loader
• Signed Installers Hide a Remote-Control Malware Network
• Poisoned AI Tools Turn Agents Into Control Planes
• Exposed Langflow Apps Become Network Pivots
• MFA Fails on the Legacy OAuth Path
• AI-Branded Extension Quietly Harvested Search Traffic
• AWS WAF Can Miss Fragmented HTTP/2 Bodies
• Bespoke Backdoor Marks CL-STA-1062’s Next Stage
• Chrome Fleet Exposure Ends Only After Every Build Updates
• AI Skills Marketplace Becomes a Credential Channel
• Cursor Sandbox Breaks on Trusted AI Inputs
• Region-Locked Banking Trojan Hides from Sandboxes
• Flowise Session Cookies Can Be Forged