🛡️ InfoSec Blue Team Briefing

Sunday, November 22, 2026

🎧 Audio Briefing

Download MP3

BlueTeamSec briefing for Sunday the 22nd of November 2026, drawn from the BlueTeamSec community on infosec.pub. It's a busy one, with 30 stories to get through, so Sebastian and I will keep each one brisk and let you decide what deserves your time.

Silent Push has uncovered a fake U.S. construction firm called The Great House, which researchers assess is a North Korean revenue operation. The operator used real-time deepfakes on video calls, and the domain registration email linked back to a list of about 1,400 known IT worker accounts. We touched on North Korean IT worker activity on the 9th, and the sloppy details, like conflicting area codes, make this a fun read.

TrendAI reports on a group calling itself BYOD, which claimed at least nine attacks in five weeks and has moved from small businesses to hospitals and universities. It gets in with stolen credentials or remote access tools rather than vulnerabilities, then goes after Active Directory and Entra. Audio ransom messages and contacting individual victims directly are the nasty extras, and it's worth a look if identity is your weak spot.

Rapid7 describes Webworm, a China-aligned espionage actor according to the researchers, spying on European governments with a rebuilt copy of the open-source SoftEther VPN Bridge rather than custom malware. They rebranded it as 'Network Monitor Software' and swapped the standard watermark for a custom one. No malware to signature, which is rather the point, so it's a good one for network-level hunting.

ReversingLabs has dug into Turla's Cereal toolkit, used against on-premises Exchange servers, and found twelve distinct loader variants that had slipped by under varied names. ESET linked the intrusions to government compromises in Kazakhstan and Romania, and the tooling evolves from the older Neuron and LightNeuron implants. If you still run Exchange on-premises, the hunting guidance is the part to read.

Amazon Web Services says Chinese-nexus actors pointed autonomous AI agents called Cairn at a list of IP addresses with the goal of getting a shell. Within 48 hours they'd found and exploited a previously unknown path traversal bug in an open-source medical imaging application, with about eleven minutes between mapping the API and writing arbitrary files. Another entry in the 'AI changes everything' file, though this one at least comes with evidence.

Palo Alto Networks looks at Com-affiliated actors running vishing campaigns since August, posing as IT support and sending staff to spoofed single sign-on portals that proxy credentials and MFA tokens. Afterwards they enrol their own MFA devices and pull data from Microsoft 365, aiming at extortion rather than encryption. The group has links to heists we saw on the 20th in Google's COLD RELIC write-up, so there's useful context there.

The U.S. Department of War has convened industry leaders around 'Cyber Effects as a Service', an outcomes-based procurement model for buying cyber operations from private firms. Industry would bring the tradecraft while the Department keeps authority over objectives and legal authorisations. Mostly a policy read, but it's interesting if you follow the commercialisation of this space.

Perplexity AI has released numbat, an open-source single-binary tool that shows defenders what AI agents are doing on endpoints, with on-device detection and optional blocking before an action runs. Worth flagging with your developers, assuming you actually know which AI tools they're using.

Microsoft has added Edge Lockdown Mode to the Canary channel, which blocks WebGL and WebGPU in the GPU process to shrink the attack surface. Microsoft estimates it would have mitigated about 60 percent of observed privilege escalation exploits since February 2022. It's aimed at high-risk users, and site-level exceptions soften the compatibility pain.

The IETF has published version eight of its Merkle Tree Certificates draft, which builds transparency logging into certificate issuance instead of bolting on separate signed timestamps. The aim is to cut certificate bloat as post-quantum signatures grow and lifetimes shrink. Certificate authorities become log operators, so it's one for the PKI nerds.

Symantec reports attackers abusing CyberPower PowerPanel Business, using its event-action feature to upload scripts through the REST API and run them as SYSTEM. They then took Active Directory data and persisted with a side-loaded DLL implant. A power management tool as an intrusion vector is a good reminder to ask what else is quietly exposed in your infrastructure.

Forescout calls it BankWeaver: attackers abused a misconfigured authentication service in Temenos Quantum Fabric banking middleware to get anonymous administrative access, then ran custom Java code and dropped web shells. It came out of a September breach at a financial institution. Another misconfiguration doing the heavy lifting, which won't surprise anyone.

SOCRadar describes BabidiGhost, a French-speaking actor who drove a jailbroken AI agent with a huge instruction file against more than 50 organisations in Belgium, France and Switzerland. Two of fourteen deeply analysed targets were fully breached, via old flaws and weak admin passwords, while MFA and IP allowlisting stopped the failures. So the AI was new, but the lesson was basic hygiene.

NOTUS reports that bank CEOs and the NSA told Congress that recent outages across aviation, banking and a classified communications network came from an aggressive AI-driven patching effort, not an attack. Organisations rushed out floods of simultaneous patches that overwhelmed legacy systems. The irony of defenders outpacing their own infrastructure is rather hard to miss.

The Associated Press says the FBI is investigating an apparent hack of the National Crime Information Center, the database law enforcement relies on. Scope, actor and vectors haven't been disclosed, and officials reportedly see no immediate link to the recent outages. Thin on detail, so one to watch rather than read.

Dragos sent nearly two million canary prompts and tokens to AI providers and found attackers had stolen data from three hosting providers, in one case intercepting requests before model processing. Stolen canary AWS keys were used for tens of thousands of API calls, and Zero Data Retention flags didn't help against insiders in provider infrastructure. Relevant if you trust AI routers and reseller marketplaces, and it follows the Cursor incident write-up from the 16th.

OX Security reports that attackers hijacked the publishing flow for Microsoft's agent-framework Python package and released malicious versions that run a payload at interpreter startup through a .pth file, with no import needed. It harvests Azure, CI, npm, Git, Docker and Hugging Face credentials, and the package gets about 192,000 monthly downloads. Treat any host that installed it as having lost every reachable secret.

Pro Concept Automotive disclosed four critical flaws, dubbed Perfekt Blue, in OpenSynergy's Blue SDK, a Bluetooth kit common in automotive and IoT, which could allow one-click remote code execution. They were found in May 2024 and patched for customers that September, with no exploitation reported. Check how long fixes take to reach the vehicles, though.

Corelight covers Ruby2Shell, a heap corruption bug in the Ruby interpreter where uppercasing the German sharp S in certain single-byte encodings expands a byte and causes an out-of-bounds write. It may be reachable remotely through web middleware like Rack, and end-of-life branches get no fix. No exploitation reported, but the detection and hunting guidance makes it useful.

A public OpenSSL GitHub issue reports that name constraints on Subject Alternative Names may not be applied when hostnames are matched against the legacy Common Name field. A certificate violating a CA's constraints could be accepted, which matters for private CAs and enterprise PKI. The source text is truncated, so treat it as a discussion to follow.

A GnuPG advisory describes a double-free in GPGME before version 2.3.1, triggered by an excessively long OpenPGP user ID after a failed buffer reallocation. It affects Gpg4win and applications using the library, such as email clients and Samba, with potential for remote code execution. No exploitation reported.

cPanel disclosed a critical flaw in how it processes SSL certificates for resellers, letting a reseller with the ssl permission run code as root. Fixes are out, and until then you can revoke that permission, even in bulk. If you host resellers, that permission audit is worth doing regardless.

The Pass-the-Device repository, released with a CODE BLUE talk, offers recon tools showing how an attacker on a compromised endpoint can spoof a trusted device against VPN and SASE products from GlobalProtect to Zscaler and Netskope. It also ships experimental Sigma rules and Sentinel queries. It undermines the comfortable zero-trust assumption, and the Netskope exploitation piece on the 20th touches the same vendor space.

DotSec found that imgmgr.exe, a Microsoft-signed utility from the Windows ADK, uses the insecure .NET BinaryFormatter, so it can run arbitrary code through a trusted binary and bypass application whitelisting. Guidance includes blocklisting it and syncing with Microsoft's recommended block rules. Several KEV entries have appeared in recent days, but this one is a pure allowlisting gap.

Coalfire's Sacrificial CoW is a Linux reflective loading trick: map a legitimate file privately, write a payload into it, and the maps entry still shows the genuine file path. That fools EDR, memory scanners and runtime monitors like Falco and Tetragon that look for anonymous executable memory, and it needs no exploit. Worth reading if your Linux detection leans on that assumption.

Hornetsecurity sees device code phishing resurging in November, using fake SharePoint payment documents to get victims to enter an attacker's code at the Microsoft device login page. That authorises a session that sidesteps passwords and MFA. Legacy auth flows left switched on again, so check whether you can restrict device code at all.

TrustedSec's Justin Elze argues that agentic pentesting frameworks make attacking low-value long-tail targets economical, so 'limited attacker attention' no longer works as a defence. He cites unauthorised access at WorkOS triggered by automated benchmark evaluations, and trojanised copies of the tools circulating. It lines up with the LLM red-teaming papers we covered on the 21st.

iVerify reports that the DarkSword iPhone exploit kit, leaked on GitHub in March, has been extended by criminal groups, including crypto wallet theft. Distribution used malvertising and abused Optimizely infrastructure, touching 178 large organisations, with compromises seen only on iOS 18.7 and earlier. Evidence fades within days and no victims received Apple threat notifications, which is rather sobering.

On a similar note, Confiant describes MoldyRoost, a malvertising campaign running from May to November that used more than 480 ad domains to deliver iOS exploit kits including DarkSword to fingerprinted iPhones in 47 countries. Cloaking let it evade ad scanners, and victims lost seed phrases, keychain data and photos. Germany's warning on mercenary spyware on the 4th is related background.

PolySwarm reports ArcticRAT, an Android RAT that now calls the xAI Grok API to triage stolen data, picking out banking apps, pulling OTPs and balances from SMS and estimating victim wealth. It spreads via malvertising and fake Play sites, and its ransomware module is fake and never encrypts. Even the malware is outsourcing its analysis now.

That's the lot for Sunday from BlueTeamSec. Where we linked activity to a country or group, that was the original authors' assessment, not ours, and the analysis was automated. Thanks for listening.

📰 Articles Covered