Latest / Neural Newscast / OpenAI GPT-5.4-Cyber and 17-Year-Old Excel Bug Risks [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] Announcer: This is Prime Cyber Insights for April 15th, 2026. [00:16] Announcer: Yesterday's Patch Tuesday was significant, and not just for the sheer volume of fixes. [00:22] Aaron Cole: We're tracking a strange intersection of the future and the past today, Aaron. [00:27] Aaron Cole: OpenAI is launching specialized defensive models, [00:31] Aaron Cole: while CISA warns about a vulnerability old enough to drive. [00:35] Announcer: Let's start with the AI frontier. OpenAI unveiled GPT-5.4 cyber yesterday, an optimized variant [00:42] Announcer: of their flagship model tailored for defensive use cases. They're also expanding the trusted [00:48] Announcer: access for cyber program to thousands of authenticated defenders. [00:52] Aaron Cole: The narrative from OpenAI is about scaling defense in lockstep with model capabilities. But Aaron, [00:59] Aaron Cole: Anthropic just previewed Mythos under Project Glasswing, which purportedly identified thousands of vulnerabilities in OS kernels and browsers. [01:08] Aaron Cole: Are defenders actually gaining the lead, OpenAI suggests? [01:12] Aaron Cole: Or is this just parity? [01:14] Announcer: It's a race, Lauren. [01:16] Announcer: OpenAI acknowledges these systems are dual use. [01:19] Announcer: The concern is that adversaries will invert these defensive models to discover zero days before they can be mitigated. [01:25] Announcer: While OpenAI highlights their Codex security agent fixing 3,000 vulnerabilities, Discovery [01:31] Announcer: is currently outstripping remediation. [01:34] Aaron Cole: That leads us to a remediation challenge CISA just flagged. [01:38] Aaron Cole: They've added CVE2009-0238 to the Known Exploited Vulnerabilities Catalog. [01:45] Aaron Cole: It's a critical RCE flaw in Microsoft Excel dating back to 2007. [01:51] Aaron Cole: Why is a 17-year-old bug surfacing in active attacks now? [01:55] Announcer: Because it still works, Lauren. [01:57] Announcer: Attackers are leveraging malformed objects in Excel documents to deliver loaders like Mdropper. [02:03] Announcer: CSAF has given federal agencies a shortened two-week deadline to patch. [02:07] Announcer: Microsoft is also adding safety rails to remote desktop files to disrupt RDP-based phishing [02:13] Announcer: used by groups like APT29. [02:15] Aaron Cole: It's a reminder that legacy tech persists indefinitely in the enterprise. [02:20] Aaron Cole: Regarding newer systems, Capsule Security just disclosed pipe leak and share leak. [02:25] Aaron Cole: These are prompt injections in Salesforce Agent Force and Microsoft Copilot that allow [02:31] Aaron Cole: attackers to exfiltrate lead databases and SharePoint data through simple lead capture [02:36] Aaron Cole: forms. [02:36] Announcer: Salesforce's response was notable. They suggested human-in-the-loop configurations to prevent unauthorized data transfers. [02:44] Announcer: It sounds like a workaround for a fundamental architectural flaw where AI agents treat untrusted inputs as trusted instructions. [02:51] Aaron Cole: Exactly. If an agent is designed for autonomy, requiring human review defeats the purpose. [02:58] Aaron Cole: We're also seeing shifts in the kernel space. [03:01] Aaron Cole: Microsoft is finally deprecating trust for cross-sign drivers this month [03:05] Aaron Cole: to disrupt the EDR killer ecosystem favored by ransomware groups. [03:09] Announcer: The BYO VD technique is becoming too effective to ignore. [03:13] Announcer: Between that and the insider threats seen with the Kraken extortion attempt this week, [03:17] Announcer: the technical perimeter is under immense pressure. [03:20] Announcer: Practitioners should prioritize identity assurance and kernel-level monitoring. [03:25] Aaron Cole: Agreed. Whether it's a 17-year-old Excel bug or a brand new GPT model, the fundamentals of least privilege and rapid patching remain the only constants. [03:35] Announcer: That's the briefing for today. [03:37] Announcer: For in-depth analysis, visit pci.neuralnewscast.com. [03:42] Announcer: Prime Cyber Insights is for informational purposes only. [03:45] Announcer: Consult with your security and legal teams before making infrastructure or policy changes. [03:50] Announcer: Neural Newscast is AI-assisted, human-reviewed. [03:53] Announcer: View our AI Transparency Policy at neuralnewscast.com. [03:58] Announcer: See you tomorrow. [03:58] Lauren Mitchell: This has been Prime Cyber Insights on Neural Newscast. [04:02] Lauren Mitchell: Intelligence for Defenders, Leaders, and Decision Makers.