🔴 CyberStrikeAI: Open-Source AI Offensive Tool Used in FortiGate Attacks Across 55 Countries
The Hacker News · AI Security
Team Cymru traced the AI-assisted FortiGate mass-compromise campaign to CyberStrikeAI, an open-source Go-based offensive platform integrating 100+ security tools with AI orchestration — built by a China-based developer with assessed government ties. First documented real-world mass campaign using an AI-native offensive security platform; this is now the threat model defenders must plan against.
🛡️ AIUC-1 Consortium: 80% of Orgs Report Risky AI Agent Behaviors, Only 21% Have Visibility
Help Net Security · AI Security
A consortium of 40+ security executives from Confluent, Elastic, UiPath, and Deutsche Börse found that shadow AI breaches cost $670K more than standard incidents, and fine-tuning attacks bypassed Claude Haiku guardrails in 72% of cases. First cross-industry quantification of the enterprise AI agent security gap — directly actionable for CISOs deploying agentic systems.
🇨🇳 DeepSeek V4 Multimodal Model Expected This Week, Optimized Exclusively for Huawei/Cambricon Hardware
TechNode · Foundation Models
DeepSeek is preparing to release V4 handling text, images, and video, timed to China's Two Sessions. The model was optimized exclusively with Huawei and Cambricon chips, deliberately withholding pre-release access from Nvidia and AMD. V4's domestic-hardware-only strategy signals an inflection point in the US-China AI supply chain divergence practitioners must plan around.
📡 More signal, less noise → www.thesignal.press