Actively exploited Security news
Security daily roundup - 21 November 2026
Saturday was about follow-ups: a critical flaw in GPGME, the library most GnuPG front ends rely on, a CVE for the bug that Cairn AI agents exploited, and a CISA deadline on Monday 23 November for exploited bugs in vCenter, Cisco Expressway and Apache Roller. Add research on a data leak from Meta's Muse assistant, Com crews vishing their way past MFA and three stories that are not yet confirmed. Twelve items worth your attention.
Key points
- Exploited bugs: CISA's deadline for vCenter, Cisco Expressway and Apache Roller is 23 November; the MONAI Label flaw exploited by Cairn AI agents is now CVE-2026-133968, with no fixed release on PyPI yet.
- Fixes and research: GPGME 2.3.1 fixes a double free triggered by a key with a long user ID, NebuSec showed a chat leak from Meta Muse, and Unit 42 detailed Com crews vishing past MFA.
- Reported, not yet confirmed: an intrusion into the FBI's NCIC database, an attempted cyberattack on the Algonquin dam and the return of the BADBOX 3.0 botnet.
Today's top stories
- vCenter: CISA's deadline is Monday. CVE-2026-123890, the zero-day Google says ARMORY CASTLE exploited, has been in the CISA KEV catalog since 20 November. US federal agencies must deal with it by 23 November, and CISA asks for forensic triage, not just patching. Our tracker lists it as a global incident. What to update and how to hunt.
- Cisco Expressway: the advisory still says nothing about attacks. Cisco's advisory for CVE-2026-76466 is still version 1.0 from 21 October, even though both security firm F6 and CISA have confirmed exploitation. The fixed release remains X15.5.20, and the CISA deadline is also 23 November. Live article.
- Apache Roller added to KEV. CISA listed CVE-2026-129448, unauthenticated code execution through XML-RPC, as exploited on 20 November, with the same 23 November deadline. Story.
- GPGME: a critical bug in the GnuPG library. An OpenPGP key with an oversized user ID triggers a double free. GnuPG says it can leak data from the application and does not rule out code execution. The fix is in GPGME 2.3.1 and Gpg4win 5.1.2, and Debian has shipped DSA-6646-1 and DLA-4924-1. Who is affected.
- Cairn: the AI-found zero-day gets a CVE. The MONAI Label flaw that, according to Amazon, Cairn AI agents found and exploited on their own against universities is now CVE-2026-133968 (CVSS 9.8). There is still no fixed release on PyPI, only a code fix. Article update.
- Meta Muse: one web page was enough. NebuSec researchers showed how a single malicious page let them send a user's chat history and saved memories from the cloud AI assistant to an outside server. They say Meta fixed the bugs before disclosure. Story.
- Com crews call as IT support. Unit 42 described attacks in which a caller steers an employee to a fake sign-in page, captures the password and MFA approval in real time and pulls data from the cloud. Finance, healthcare and law firms are the main targets. How to defend.
- Signal pushes back on phone AI assistants. On iOS 27 it blocks screenshots because of Siri's on-screen awareness, and on Android it turns on screen security when an AI assistant tries to read the screen. Signal itself says the protection is incomplete. Story.
- Hoax threats against Czech schools go to court. Ukraine's cyberpolice say the case against a man from Dnipro, who had sent hoax bomb threats to schools in Czechia, Slovakia and Latvia since September 2024, has been sent to court. Cyberpolice statement (in Ukrainian).
- Apify closes its incident. On 20 November the company marked the 9-10 November breach as resolved and published a preliminary report. A full report will follow the investigation; our advice to rotate tokens still stands. Update.
- FBI NCIC and the Algonquin dam (reported). According to CNN, the FBI is investigating an intrusion into the database US police use to share warrants, and the FBI has not commented publicly (what we know). According to an Illinois DNR statement quoted by ABC7, the Algonquin dam faced an attempted cyberattack on 16 November and kept operating normally (story).
- BADBOX 3.0 (reported). According to BleepingComputer, the botnet living in cheap Android TV boxes and tablets is back in a new version; HUMAN's Satori team puts its peak at up to 40 billion fraudulent ad requests a day. How to spot an infected device.
Still open
- NGINX2Root: since 21 November our tracker lists it only as ELEVATED. Nobody has shown exploitation in the wild yet, and Ubuntu, Debian 12 and Debian 13 all have fixed packages.
- PHP: the six critical bugs from the 19 November release stay at ELEVATED, with no reported exploitation so far.
- FBI NCIC and Algonquin are waiting for confirmation from primary sources - we will upgrade or correct those stories when it arrives.
What to do today
- Patch vCenter (VMSA-2026-0009), Cisco Expressway (X15.5.20) and Apache Roller, then check logs for signs of exploitation.
- Update GPGME to 2.3.1, Gpg4win to 5.1.2 on Windows, and on Debian install the packages from DSA-6646-1 and DLA-4924-1.
- If you run MONAI Label, keep it off the internet until a fixed release ships.
- Remind staff and the help desk that IT never asks for an MFA code over the phone.
Sources
Spotted a mistake? Let us know and we will correct it openly.