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
• Microsoft 365 Phishing Now Leaves Long-Lived Control
• 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

Recent editions

• Tuesday, June 30, 2026: Oracle Payments Bug Can Leak the Keys Behind It · 17 stories
• Monday, June 29, 2026: Oracle Payments Bug Can Leak the Keys Behind It · 13 stories
• Sunday, June 28, 2026: Recovery Keys Let Messaging Account Theft Persist · 1 story
• Saturday, June 27, 2026: Repo Opens Became Live Cloud Compromise · 14 stories
• Friday, June 26, 2026: Bespoke Backdoor Marks CL-STA-1062’s Next Stage · 17 stories
• Thursday, June 25, 2026: Internet-Facing PeopleSoft Console Becomes Extortion Foothold · 17 stories
• Wednesday, June 24, 2026: Internet-Facing PeopleSoft Console Becomes Extortion Foothold · 19 stories