Latest / Neural Newscast / Industrialized Social Engineering and Flowise AI Exploitation [Prime Cyber Insights]
Transcript
- [00:00] Announcer: From Neural Newscast, this is Prime Cyber Insights, [00:03] Announcer: Intelligence for Defenders, Leaders, and Decision Makers. [00:11] Aaron Cole: Welcome to the Briefing Room. Today, April 7th, 2026, we are analyzing a fundamental shift in [00:18] Aaron Cole: threat actor tradecraft, the industrialization of social engineering. From the Axios NPM compromise [00:26] Aaron Cole: to the drift protocol heist, attackers are moving beyond simple phishing into long-term, [00:32] Aaron Cole: high-trust operations. Aaron, it's... [00:34] Lauren Mitchell: And it's not just the trust building that's evolving. [00:37] Lauren Mitchell: It's the target selection. [00:39] Lauren Mitchell: The Axios attack saw a lead maintainer, Jason Simon. [00:43] Lauren Mitchell: Again, targeted for two weeks by North Korean group UNC1069, using a fake Microsoft Teams [00:51] Lauren Mitchell: update. [00:51] Lauren Mitchell: They aren't just looking for one wallet anymore. [00:54] Lauren Mitchell: They're looking for right access to packages with 100 million weekly downloads. [00:59] Aaron Cole: Exactly. [01:01] Aaron Cole: We're seeing this same slow-burn methodology in the Drift Protocol hack. [01:05] Aaron Cole: It wasn't a one-click exploit. [01:07] Aaron Cole: The attacker spent six months meeting contributors in person at crypto conferences, posing as a quantitative trading firm. [01:14] Aaron Cole: They eventually stole over $280 million by hijacking the Security Council's administrative powers. [01:21] Lauren Mitchell: Erin, that in-person element is a significant escalation for practitioner threat modeling. [01:27] Lauren Mitchell: While we're looking for malicious code, we also have to account for human-centric reconnaissance [01:33] Lauren Mitchell: that bypasses standard 2FA once the initial device is compromised. [01:37] Lauren Mitchell: It effectively renders traditional identity controls moot once a rat is established on the administrator's machine. [01:45] Aaron Cole: Shifting to the infrastructure side, FlowEye's AI is facing a critical situation. [01:51] Aaron Cole: Volnczek is reporting active exploitation of CVE 2025-5928A. [01:57] Aaron Cole: This is a CVSS 10.0 code injection flaw in their AI agent builder. [02:02] Aaron Cole: There are over 12,000 exposed instances online right now, and the exploit is coming from a single Starlink IP address. [02:10] Lauren Mitchell: The technical risk here is severe, Aaron. [02:13] Lauren Mitchell: FlowEyes noted the flaw allows access to Node.js runtime privileges, [02:17] Lauren Mitchell: specifically the child process and file system modules. [02:21] Lauren Mitchell: It means full system compromise. [02:24] Lauren Mitchell: If you're running FlowEyes, you must move to version 3.0.6 immediately. [02:29] Lauren Mitchell: This follows a pattern of AI management tools becoming a primary entry point for RCE. [02:35] Aaron Cole: The velocity of these attacks is also increasing. [02:38] Aaron Cole: Microsoft Threat Intelligence just linked the China-based Storm 1175 to a chain of zero-day exploits used to deploy Medusa ransomware. [02:47] Aaron Cole: In some cases, they are moving from initial access to full data exfiltration in under 24 hours, often targeting healthcare and finance. [02:55] Lauren Mitchell: Storm 1175 is rotating exploits faster than most organizations can patch, Aaron. [03:01] Lauren Mitchell: They've exploited 16 vulnerabilities since 2023, including the recent SmarterMail and Beyond Trust flaws. [03:10] Lauren Mitchell: They are leveraging living off the land binaries like PowerShell and Impactit to blend in, [03:16] Lauren Mitchell: making the detection window for a 24-hour ransomware deployment incredibly narrow. [03:22] Aaron Cole: We also need to look at hardware-level risks. [03:25] Aaron Cole: A University of Toronto team just revealed GPU breach. [03:28] Aaron Cole: This is a Rohammer-style attack on GDDR6 memory [03:32] Aaron Cole: that enables root shell access on systems like the NVIDIA RTX A6000. [03:37] Aaron Cole: It's particularly dangerous because it doesn't require disabling IOMMU protection. [03:43] Lauren Mitchell: That's the critical takeaway for our technical audience, Aaron. [03:46] Lauren Mitchell: IOMMU was the hardware unit meant to stop these direct memory attacks. [03:51] Lauren Mitchell: GPU breach shows that even with it enabled, a malicious CUDA kernel can corrupt GPU page tables to gain read-write access. [04:00] Lauren Mitchell: It basically breaks the hardware security boundary for consumer GPUs that lack error-correcting [04:07] Lauren Mitchell: code memory. [04:08] Aaron Cole: Lastly, the BrowserGate report claims LinkedIn is secretly scanning user browsers for over [04:14] Aaron Cole: 6,200 installed extensions, allegedly to map competitor usage. [04:18] Aaron Cole: Whether it's for policy enforcement or corporate espionage, it's a reminder that our primary work tools are also significant telemetry collectors. [04:27] Lauren Mitchell: It's a high-stakes landscape, Aaron. [04:30] Lauren Mitchell: From AI-assisted supply chain scans on GitHub to in-person social engineering, the common thread today is scale and sophistication. [04:39] Lauren Mitchell: That covers our priority updates for this session. [04:42] Aaron Cole: Thanks, Lauren. [04:43] Aaron Cole: For further technical details, check the documentation at pci.neuralnewscast.com. [04:49] Aaron Cole: This podcast is for informational purposes only and does not constitute professional security advice. [04:55] Aaron Cole: Neural Newscast is AI-assisted, human-reviewed. [04:58] Aaron Cole: View our AI transparency policy at neuralnewscast.com. [05:02] Aaron Cole: Stay focused. [05:03] Announcer: This has been Prime Cyber Insights on Neural Newscast, Intelligence for Defenders, Leaders, and Decision Makers.