Latest / Neural Newscast / AI Chatbot Poisoning and Megalodon GitHub Attacks [Prime Cyber Insights]
Transcript
- [00:00] Announcer: From Neural Newscast, [00:01] Announcer: this is Prime Cyber Insights Intelligence for defenders, leaders, and decision makers [00:11] Announcer: Welcome to the briefing room at Prime Cyber Insights. Today is May 27th, 2026. [00:18] Announcer: We're tracking a fundamental shift in how attackers reach users, [00:22] Announcer: moving beyond traditional search engines and into the AI ecosystem. [00:26] Aaron Cole: Our focus today [00:27] Lauren Mitchell: starts with Microsoft's warning that AI chatbots are now being used to recommend [00:32] Lauren Mitchell: malicious software download sites, [00:34] Lauren Mitchell: essentially extending SEO poisoning into large language model interactions. [00:40] Aaron Cole: Exactly, Lauren. Microsoft released a report Tuesday detailing a campaign where chatbots [00:45] Aaron Cole: when queried for software recommendations provided links to domains like gleese.com. [00:51] Aaron Cole: These sites impersonate utilities like Firmark or Crystal Disk Info. The end goal is crypto-jacking [00:58] Aaron Cole: but the delivery is sophisticated using screen connect for persistence [01:02] Aaron Cole: and targeting systems with high performance GPUs to maximize mining yield. [01:06] Lauren Mitchell: It is a deliberate strategy, Aaron. By focusing on mining value rather than infection count, [01:12] Lauren Mitchell: they are being more efficient. We are also seeing a massive click-fix campaign exploiting [01:18] Lauren Mitchell: a critical SQL injection flaw in Ghost CMS. Tracta's 2026 Cape Verde Escudos 26980 Researchers [01:27] Lauren Mitchell: from Chianzine and MalwareBytes note it is hitting over 700 domains, including Harvard and Oxford, [01:33] Lauren Mitchell: tricking users into running PowerShell commands disguised as Cloudflare verifications. [01:39] Aaron Cole: The scale of that Ghost CMS campaign is alarming, Loren, but it is being matched by automation [01:45] Aaron Cole: in the developer space. This week we analyzed the Megalodon campaign. On May 18th, [01:50] Aaron Cole: in just a six-hour window, [01:52] Aaron Cole: a threat actor pushed over 5,700 malicious commits to more than 5,500 GitHub repositories. [01:59] Lauren Mitchell: That speed is incredible, Aaron. Safety reported these commits exfiltrate CI CD secrets [02:06] Lauren Mitchell: and SSH keys. There is speculation about a connection to Team PCP, the group that recently claimed [02:13] Lauren Mitchell: a major breach of GitHub's internal repositories, though forensic links [02:17] Lauren Mitchell: are still being verified. It reinforces the danger of automated supply chain poisoning. [02:22] Aaron Cole: And the barrier to entry for these AI-related attacks is dropping, Lauren. [02:27] Aaron Cole: A tool called Heretic was recently highlighted for its ability to obliterate [02:31] Aaron Cole: or strip AI guardrails from open-source models like Llama-3 in under 10 minutes, [02:37] Aaron Cole: allowing them to provide instructions for everything from ricin dosages to credit card theft. [02:42] Lauren Mitchell: The technical evolution is matched by geopolitical persistence. Researchers at Gambit Security [02:49] Lauren Mitchell: now say the March breach of the Los Angeles transit system was the work of Iranian backhackers [02:53] Lauren Mitchell: linked to the Ministry of Intelligence and State Security. The group, [02:57] Lauren Mitchell: calling itself Ababil of Menab, claims they deleted data during the multi-week incident. [03:03] Aaron Cole: On the enforcement side, Dutch police announced the arrest of a 35-year-old [03:07] Aaron Cole: suspect for the early 2026 hack of the Ajax football club. That attacker reportedly had the ability [03:13] Aaron Cole: to view details on 300,000 accounts and even manipulate stadium bans. [03:17] Aaron Cole: It is a rare win in a landscape that is changing faster than most perimeters can handle. [03:22] Lauren Mitchell: That brings us to the bigger picture, Aaron. As Dark Reading notes in their 20th anniversary [03:27] Lauren Mitchell: retrospective today, we have moved from simple perimeter defense to an era of identity [03:33] Lauren Mitchell: needed security. The shift from hardware appliances to software and service-based security [03:38] Lauren Mitchell: is not just a trend; it is a requirement to handle these distributed AI-assisted threats. [03:43] Aaron Cole: Direct action for practitioners, upgrade [03:46] Lauren Mitchell: GhostCMS to version 6.19.1 immediately and audit your GitHub action workflows for [03:51] Lauren Mitchell: any SysTag YAML files. For more technical deep dives, visit pcineuralnewscast.com. [03:57] Lauren Mitchell: Thank you for joining us in the briefing room, Lauren. Stay resilient, Aaron. Remember, [04:01] Lauren Mitchell: the information in this briefing is for educational purposes only. This is Prime Cyber Insights from [04:07] Lauren Mitchell: Neural Newscast. Neural Newscast [04:09] Lauren Mitchell: is AI-assisted human-reviewed. View our AI transparency policy at neuralnewscast.com. [04:15] Announcer: This has been Prime Cyber Insights on Neural Newscast. [04:19] Announcer: Intelligence for defenders, leaders, and decision-makers.