Latest / Neural Newscast / CISA GitHub Leak Lessons and the FSB Router Threat [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: I'm Aaron Cole. This is Prime Cyber Insights, professional analysis for the security practitioner. [00:18] Aaron Cole: We're starting today with CISA's blunt post-mortem on its own internal security failure. [00:24] Aaron Cole: Yesterday, the agency detailed how a contractor published 844 megabytes of sensitive data to a public GitHub repository titled Private CISA. [00:34] Aaron Cole: The leak included administrative keys for three AWS GovCloud servers and a CSV file containing plain text passwords for dozens of internal systems. [00:44] Aaron Cole: Lauren, the timeline on this is what really bothers me. [00:47] Lauren Mitchell: It should, Aaron. That repository was public for nearly six months before Krebs on Security flagged it. [00:55] Lauren Mitchell: Even after the alert on May 15th, CISA admitted it took more than 48 hours to invalidate the AWS keys. [01:03] Lauren Mitchell: Their explanation cites complexities and interconnections with federal partners that slowed the rotation. [01:09] Lauren Mitchell: For a practitioner, the takeaway is clear. [01:12] Lauren Mitchell: If your key rotation playbook isn't well-tested for cloud services, [01:16] Lauren Mitchell: the rotation will fail under the pressure of a live incident. [01:19] Announcer: And it wasn't just a technical lag. [01:22] Aaron Cole: CISA ignored nine automated alerts from GitGuardian prior to the manual notification. [01:28] Aaron Cole: Acting CIO Preston Wurntz and Acting CISO Brad Libby noted that reporting channels were poorly defined, [01:34] Aaron Cole: causing the researcher to cycle through multiple avenues before involving the media. [01:38] Aaron Cole: They're now advocating for better use of the security.txt standard, [01:43] Aaron Cole: but the internal friction suggests CISA's vulnerability disclosure platform [01:46] Aaron Cole: wasn't ready to handle its own infrastructure leaks. [01:50] Lauren Mitchell: Exactly. [01:51] Lauren Mitchell: Moving from internal leaks to external state-sponsored threats, the United States government, alongside international allies, issued a massive advisory this week regarding Russia's FSB Center-16. [02:03] Lauren Mitchell: Groups like Berserk Bear and Ghost Blizzard are aggressively targeting home and small office routers. [02:09] Lauren Mitchell: They aren't looking for data on those devices. They're turning them into an exit node botnet to hit critical infrastructure. [02:17] Aaron Cole: Ars Technica notes they're specifically scanning for SNMP agents that accept default authentication. [02:22] Aaron Cole: Once in, they use the SNMP agent to copy the device configuration to an attacker-controlled server. [02:28] Aaron Cole: The goal is to use residential IPs as proxies, lowering the chance that defense or energy sector firewalls will block their traffic. [02:35] Aaron Cole: The fix is technically simple, but hard to enforce at scale. [02:39] Aaron Cole: Disable SNMP versions 1 and 2 entirely. [02:42] Aaron Cole: Which brings us to a more modern infrastructure risk. [02:46] Aaron Cole: Varonis researchers recently disclosed a critical vulnerability in Google Cloud's Dialogflow [02:50] Aaron Cole: CX, as reported by TechRadar. [02:53] Aaron Cole: By exploiting code blocks in conversation playbooks, researchers created rogue agents [02:58] Aaron Cole: that could exfiltrate chat history and credentials. [03:01] Aaron Cole: Because these agents share a cloud-run environment within a project, one compromised agent could take over every other agent in that same project. [03:10] Aaron Cole: The detection gap there is significant, Lauren. [03:12] Aaron Cole: Standard cloud logging doesn't capture the file overwrites or the injected logic used in these AI agents, making the attack virtually undetectable. [03:21] Aaron Cole: Google patched the issue between April and June, but researchers are still advising practitioners to manually inspect code blocks for unauthorized Python snippets and review audit logs for anomalous session errors. [03:31] Lauren Mitchell: It's a reminder that as we rush to adopt AI agents, we're creating shared execution environments that often lack the granular visibility we've spent years building for traditional web apps. [03:43] Lauren Mitchell: Whether it's router hygiene, GitHub secret management, or AI platform configurations, [03:49] Lauren Mitchell: the fundamental issue remains credential and identity governance. [03:53] Announcer: Agreed. I'm Aaron Cole. Stay sharp and stay direct. [03:58] Lauren Mitchell: And I'm Lauren Mitchell. [03:59] Lauren Mitchell: Prime Cyber Insights is a production of the Neural Newscast Network. [04:03] Lauren Mitchell: For technical deep dives into these stories, visit pci.neuralnewscast.com. [04:08] Lauren Mitchell: This briefing is provided for informational purposes only and does not constitute legal or technical advice. [04:14] Lauren Mitchell: Neural Newscast is AI-assisted, human-reviewed. [04:17] Lauren Mitchell: View our AI transparency policy at neuralnewscast.com. [04:21] Lauren Mitchell: We'll see you in the briefing room tomorrow. [04:23] Announcer: This has been Prime Cyber Insights on Neural Newscast. [04:27] Announcer: Intelligence for defenders, leaders, and decision makers.