Latest / Neural Newscast / Five Eyes Warns of AI Cyber Attacks Within Months [Prime Cyber Insights]
Transcript
- [00:00] Announcer: From Neural Newscast, this is Prime Cyber Insights, intelligence for defenders, leaders and decision makers. [00:10] Aaron Cole: Lauren, let's start with that warning from the Five Eyes. [00:14] Aaron Cole: Earlier this week, the Alliance, comprising the United States, United Kingdom, Canada, [00:18] Aaron Cole: Australia, and New Zealand issued a direct alert. [00:21] Aaron Cole: They anticipate Frontier AI models will transform offensive capabilities within months, [00:27] Aaron Cole: specifically pointing to models like Anthropik's Mythos Preview, [00:30] Aaron Cole: which Mozilla has confirmed is powerful enough to require restricted access. [00:35] Lauren Mitchell: The language in that statement was remarkably direct, Aaron. [00:38] Lauren Mitchell: They are no longer treating this as a technical curiosity. [00:42] Lauren Mitchell: By framing cyber resilience as integral to business continuity and long-term value, [00:47] Lauren Mitchell: they are moving the responsibility from the server room straight to the boardroom. [00:52] Aaron Cole: Speaking of the boardroom, Meta is currently navigating significant internal backlash. [00:56] Aaron Cole: They have paused their Model Capability Initiative, or MCI, which was designed to train AI systems by logging employee keystrokes, mouse movements, and screen content. [01:05] Aaron Cole: Reports from Wired indicate an internal security review found this granular data was accessible to thousands of staff members across thousands of internal data tables. [01:14] Lauren Mitchell: That is a massive exposure of high-risk data, Aaron. [01:18] Lauren Mitchell: We're talking about transcriptions, private conversations, and authentication flows. [01:23] Lauren Mitchell: Meta's CTO previously confirmed there was no opt-out for United States workers on company devices. [01:29] Lauren Mitchell: Following an internal security notice on June 18th, they have locked down the MCI database entirely while investigating the scope of improper access. [01:37] SPEAKER_03: It is a stark reminder that if you collect it, you must protect it. [01:42] SPEAKER_03: Moving to the software supply chain, we are looking at a severe compromise of Shaped Plugins Pro WordPress tools. [01:48] SPEAKER_03: WordFence reported that attackers poisoned the official distribution pipeline for plugins like Product Slider Pro and Real Testimonials Pro. [01:55] SPEAKER_03: This was a targeted hit on the vendor's paid infrastructure, rather than the free versions available on WordPress.org. [02:01] Lauren Mitchell: The payload is sophisticated. [02:04] Lauren Mitchell: It triggers a loader on admin pages that installs a hidden plugin capable of capturing plain text credentials and two-factor codes. [02:11] Lauren Mitchell: It even targets WooCommerce order data and SMTP credentials. [02:15] Lauren Mitchell: For site owners, the 2026 Cape Verde Escudo's 49,7777 identifier carries a CVSS score of 10.0. [02:24] Lauren Mitchell: If you're running these pro versions, you must check for versions released before 3.5.4 or 4.0.2 immediately. [02:32] Aaron Cole: On the automated front, the Jared from Subway MEV bot just took a $15 million hit. [02:38] Aaron Cole: Blockade detected that an attacker used fake cryptocurrency pools and tokens to trick the bot's logic into granting spending approvals. [02:45] Aaron Cole: It was a calculated heist where the attacker accumulated permissions over multiple test transactions before draining the WETH and United States DC. [02:53] Aaron Cole: It demonstrates that even the most aggressive automated systems have logical blind spots. [02:58] Aaron Cole: Finally, Aaron, Microsoft addressed the AutoJack vulnerability chain in Autogen Studio. [03:04] Aaron Cole: While caught before a wide release, it highlighted how a malicious web page could trick an AI agent into executing arbitrary commands on a developer's host system. [03:12] Aaron Cole: That is why the recommendation remains to run these agentic AI frameworks [03:16] Aaron Cole: in strictly isolated, low-privilege environments. [03:19] Aaron Cole: That wraps up our briefing for today. [03:21] Aaron Cole: I'm Erin Cole. [03:22] Aaron Cole: And I'm Lauren Mitchell. [03:23] Aaron Cole: For the latest technical deep dives and risk analysis, visit PCI.NeuralNewscast.com. [03:28] Aaron Cole: Neural Newscast is AI-assisted, human-reviewed. [03:31] Aaron Cole: View our AI transparency policy at NeuralNewscast.com. [03:34] Aaron Cole: We'll see you in the briefing room tomorrow. [03:36] Announcer: This has been Prime Cyber Insights on Neural Newscast, [03:40] Announcer: Intelligence for Defenders, Leaders, and Decision Makers