A patch that's been out for three weeks, and more than a thousand GitLab servers still haven't installed it. Now attackers are using that flaw. Welcome to Lantern Cybersecurity Daily. It's Tuesday, November 3rd. Today: actively exploited GitLab servers, an AI agent framework CrowdStrike ties to a campaign against more than two hundred organizations, and Google cutting off an Android TV box maker it says built the malware in at the factory. Hit follow and you won't have to come looking for the next episode.
BleepingComputer reports:
On Friday, the US Cybersecurity and Infrastructure Security Agency (CISA) confirmed that attackers have begun exploiting the flaw in the wild and added CVE-2026-97229 to its Known Exploited Vulnerabilities (KEV) Catalog, ordering federal agencies to secure their servers by Sunday. Although GitLab released patches three weeks ago, threat watchdog Shadowserver revealed over the weekend that it tracks more than 1,000 unpatched GitLab servers exposed to ongoing attacks.
So the people affected are teams running self-managed GitLab, Community or Enterprise Edition. The bug sits in the regular expression compiler that reads CI/CD pipeline configs, and GitLab says that under certain conditions an unauthenticated user could run arbitrary code on the server. GitLab.com is already patched, and Dedicated customers don't need to do anything. Which means the thousand holdouts are mostly someone's GitLab box in a closet. This one does justify interrupting the week, though. Plenty of CVEs don't, but this has a named product, CISA confirming exploitation, and fixed versions you can install: 19.4.2, 19.3.5 and 19.2.8, released October 12th. And think about what's on a GitLab server: source code, pipeline secrets, deploy credentials. Shadowserver also counts more than eleven thousand instances still open to a different maximum-severity regex bug, CVE-2026-110264, out of the roughly thirty-nine thousand it sees each day. CISA hasn't flagged that one as exploited in this reporting. So there are two separate questions. If you upgrade today, you close the hole. But your server was exposed for three weeks while a fix existed, so you also have to check whether anyone got in before you closed it.
Next, BleepingComputer on new CrowdStrike research:
The AI agents performed reconnaissance, exploited vulnerabilities, deployed web shells, obtained credentials, and pursued alternative attack paths when needed. The operators explicitly instructed the agents to skip Windows machines protected by Falcon, CrowdStrike's solution for endpoint protection, and focused on unprotected hosts.
CrowdStrike attributes this to an actor it tracks as Vapor Panda, and calls it China-nexus. That attribution is CrowdStrike's. According to the researchers, the operation started October 17th, went after more than two hundred organizations in fourteen countries, and compromised at least nine. The confirmed victims they list include telecoms in the US and Belgium and US pharmaceutical firms. And of course the vendor's own write-up says the attackers steered around the vendor's product. I'm not saying it's false, just that you should notice who's telling it. What I find more interesting is the method. CrowdStrike says a human operator gave instructions in Chinese and stepped in at decision points, while ARTEX agents ran several intrusions at the same time. One intrusion burned through more than 1.4 billion input tokens, by the researchers' estimate. Here's the detail that matters for defenders. Against a Greek government entity, CrowdStrike says, the agents forged an administrator token by using authentication keys from a mail-server directory export dating to 2022. The AI didn't need a new trick there. It found old keys nobody had rotated. That fits what the Dutch Institute for Vulnerability Disclosure now says about its own September breach. It says ARTEX went from initial access to root in twelve minutes by chaining two zero-days, and called the activity super loud and messy. Four hundred fifty thousand requests is a lot of noise. Your logs should be able to catch that, if anyone is watching them.
And BleepingComputer on a botnet takedown by Google:
"We assess with high confidence that the backdoor and botnet were intentionally developed, deployed, and operated for financial gain by Videostrong Technology Co.," Google's Threat Intelligence Group (GTIG) said in a report published Monday.
The botnet, XCPE, covers more than 1.3 million Android TV streaming boxes. Google says the infection wasn't a vulnerability or a malicious download. The backdoor was in the firmware from manufacturing and ran with system-level privileges. Google accuses Videostrong, a Chinese manufacturer, of selling access to criminal affiliates, who used it for ad fraud and residential proxies and to install more malware. That's Google's assessment, not a court finding. The consequences are concrete, though. Google revoked Videostrong's license to make Android TV devices with Google apps and banned it, and any future affiliates, from certified Android. Working with Japan's National Police Agency and companies including Lumen and HUMAN Security, Google also took down command servers used by more than a dozen related malware families. Google says that cut off at least ten million more compromised devices, including set-top boxes, karaoke machines and telehealth hardware, and not all of them run Android. Your network can be full of devices like that, and nobody thinks of them as computers. And there's nothing the owner can patch. If the factory put the backdoor there, the best you can do is keep a cheap box on its own network segment, away from anything you care about. Taking down the servers protects people for now. It doesn't remove the code from the firmware.
If the ARTEX story is your kind of thing, try our sister show, AI Daily Briefing. Every weekday it covers the top AI news for engineers, founders and investors, and separates real capabilities from demo hype quickly.
To sum up: upgrade self-managed GitLab and look for signs of an earlier intrusion, rotate keys that should have expired years ago, and give your cheap streaming boxes their own network. We'll be back tomorrow. Lantern Cybersecurity Daily is a Lantern Podcast, and thanks for listening.