A second sign-in method left switched on can quietly undo multi-factor authentication on a SonicWall VPN. Attackers have already used that gap. This is Lantern Cybersecurity Daily, with a Saturday edition. After SonicWall: home and small-office storage boxes running Plex, two new looks inside AI-driven attack operations, and OpenAI's most capable vulnerability-hunting model yet. Let's begin at the VPN.
BleepingComputer, on the SonicOS advisory:
The vulnerability, tracked as CVE-2026-19868, allows attackers with valid credentials for an alternate authentication method to bypass SAML authentication controls, potentially including SAML-enforced multi-factor authentication (MFA). The flaw has a CVSS severity score of 7.1 and only affects SonicOS SSL-VPN services configured to use SAML authentication.
A 7.1 doesn't sound like a weekend problem. But this is the one where the severity score undersells it: CISA added it to its Known Exploited Vulnerabilities catalog on November sixth, and federal agencies have until November twentieth. The affected group is specific. Organizations running SonicOS SSL-VPN with SAML handling sign-in, often to enforce MFA, where some other way to authenticate is still enabled. An attacker with working credentials for that other method can get past the SAML controls. SonicWall says it investigated a report of potential exploitation but hasn't shared details; CISA's listing is what confirms it's happening in the wild. For a small team, the triage is short. Gen7 at 7.3.3-7015 or earlier, or Gen8 at 8.2.2-8015 or earlier, is affected. Fixes are 7.3.4-7013 and 8.2.3-8015, and SonicWall says there are no workarounds. Advisory firm Black Shard adds a sensible first look: list every non-SAML sign-in method on each device and switch off what nobody uses. The same update fixes six more SonicOS flaws, including two management-interface bugs rated 8.8. And this follows months of SonicWall SMA1000 attacks. Customers deserve an account of what that exploitation report actually showed.
From BleepingComputer, on QNAP's advisory:
QNAP says that users should treat their NAS devices and all personal content as compromised if they're running a vulnerable Plex server that is exposed on the public web. The vendor warned that updating now will not remove any payloads the attackers have already installed on the device.
That's an unusually blunt vendor sentence, and I appreciate it. If you run Plex Media Server on a QNAP NAS and it's reachable from the internet, an update is not the finish line. QNAP says some users report Plex on their NAS appears compromised through Plex2Shell, a chain of four Plex Media Server flaws. Full technical details went public on October sixteenth, exploitation followed shortly after, and CISA added all four to its exploited list on October twentieth. QNAP's own operating systems, it says, aren't affected; the hole is in the Plex app. So the work for an exposed box: check QNAP's indicators of compromise, do a clean reinstall of QTS or QuTS hero, then reset Plex and NAS credentials. If you don't need remote access, stop forwarding port 32400. If you do, QNAP says go through a VPN. Backups and media on one appliance, now with a reinstall on the to-do list. Fair, but the vendor is doing the useful thing here: a step-by-step remediation guide and an offer to review submitted logs. The current Plex build in the App Center is 1.43.3.10861.
Check Point Research, published November fifth:
As described further below, artifacts obtained by Check Point Research show that these systems can work fully autonomously after the initial goal has been set by the operator. Using agentic hacking frameworks like ARTEX, threat actors can carry out attacks at a previously impossible speed, on a broad scale and without any technical knowledge, opening the field to a large set of people.
We've followed ARTEX since CrowdStrike tied it to a China-nexus campaign on November third, and WorkOS blamed it for its sales-app breach the next day. This is the criminal side, seen from the operators' own consoles. Check Point found ten ARTEX instances misconfigured to expose their full databases. And the operator chat is mostly "hello" and "what's the progress." In one campaign, agents went after ten gray-market iOS signing shops, pulled more than 300 certificates with passwords, and validated three as usable. Check Point estimates the model bill at somewhere between 700 and 1,500 dollars. That's a low price for an operator who mostly typed hello. Microsoft's report the same day matters more for cloud teams. It says an agentic framework searched public GitHub for exposed secrets, and in one case went from scanning to pulling data out of Azure Key Vault in five minutes. Microsoft also describes service-principal sign-ins with stolen app credentials from October twenty-eighth, from IP addresses it says Todyl also reported in the campaign we covered Thursday. Microsoft's advice is concrete: deleting the secret from code doesn't fix anything until it's rotated or revoked, and it recommends moving apps off client secrets to managed identities. That's the ticket worth filing Monday.
BleepingComputer, on OpenAI's new model:
OpenAI has classified the new model as having "Critical" cybersecurity capabilities under its Preparedness Framework, up from the "High" classification given to its previous model. The new model is not publicly available, with access currently restricted to a select group of approved customers in OpenAI's Daybreak Red.
OpenAI announced GPT-6 Cyber on Thursday with some big numbers: 81 previously unknown high-severity Firefox vulnerabilities, and working local privilege escalation exploits for 948 distinct Linux kernel flaws. Those are OpenAI's figures, from OpenAI's testing. Right, and there's one piece of outside evidence: many of the Firefox patches have already been released by Mozilla. For builders, the consequence is the pace of fixes. OpenAI itself says defenders have a limited window before similar capability is more widely available to attackers. This week's OpenSSH login flaw was also found with AI agents running about 17,600 authentication tests. My worry is the receiving end. A model that finds bugs by the hundred lands them on maintainers who fix them by the handful. OpenAI says it plans to widen its Daybreak programs to open-source maintainers and independent researchers; that's the part I'd watch. Agreed. Access control on the model is OpenAI's decision. Patch capacity downstream is everyone's problem.
One more listen for your queue. Plenty of today's attack surface lives in the cloud, and if you want the infrastructure underneath it, try The Data Center Daily: a daily briefing on AI compute, hyperscaler capex, the power grid, semiconductor supply, and energy markets reshaped by intelligence at scale. It's wherever you get podcasts.
Before the weekend gets away: if you run SonicWall SSL-VPN with SAML, or Plex on an internet-facing NAS, those are the checks to make first. The sources for every segment are in the show notes. We're back Monday, November ninth. Thanks for spending part of your Saturday with Lantern Cybersecurity Daily, which comes to you as a Lantern Podcast.