CyberPulse

The Agent Internet Is Already Leaking

Board-level strategic briefing on the defining risk of 2026: AI agents are being deployed across enterprises as autonomous actors with credentials, permissions, and action authority — but without the identity governance applied to human employees. Gartner projects 40% of enterprise applications will embed AI agents by year-end 2026 (up from 5% in 2025). 92% of security leaders are concerned about...

North Korea's ScarCruft built Ruby Jumper, a five-component toolchain that breaches air-gapped networks by installing a disguised Ruby runtime, weaponizing USB drives as bidirectional command channels, and deploying full-spectrum surveillance including keylogging, audio, and video capture inside physically isolated environments. Separately, Aeternum C2 is a new botnet that writes encrypted command...

A rival AI company officially launched GPT-5.4-Cyber, a frontier model optimized for defensive cybersecurity, days after Project Glasswing — making the AI cyber arms race live with two competing frontier models now shipping for enterprise security. A critical authentication bypass (CVE-2026-33032, CVSS 9.8) was discovered in the Model Context Protocol (MCP) integration for nginx-ui, allowing any n...

Check Point Research disclosed critical vulnerabilities in Anthropic's Claude Code where simply opening an untrusted repository could silently execute commands on a developer's machine, steal API credentials, and compromise an entire team's workspace — all through configuration files treated as harmless metadata. Separately, Google and Mandiant dismantled GRIDTIDE, a China-linked espionage campaig...

The Criminal Marketplace Youre Not Monitoring

The Death of Human Error