N-central Zero-Days, Outlook NTLM Theft, Apache HTTP/2 RCE, & AI Agents Leaking Secrets
Patched Servers, Stolen Hashes, & Secrets in Plain Sight
Attackers got past this month’s emergency fixes, pulled credentials from a single opened file, and went looking for secrets that AI tools left in public code.
This week, patched did not mean protected. Attackers broke into N-able N-central servers that already had the vendor’s emergency fix, and Microsoft patched an Outlook zero-day that leaks a user’s Windows sign-in hash. Apache fixed a critical web server flaw that Anthropic and Google both reported.
Meanwhile, researchers found an AI coding assistant saving passwords and API keys into project files that end up in public GitHub repositories, where attackers already hunt for credentials. Together, the stories show that installing an update is where the work starts, not where it ends.
At a Glance
N-able N-central Zero-Days Hit Servers That Were Already Patched
Attackers compromised N-central servers within hours of a new exploit going public, and most of those servers already had N-able’s emergency fix. N-central is the remote monitoring and management (RMM) platform many MSPs and IT teams use to run their clients’ endpoints.
From disclosure to compromise
- Nov 8Zero-days go public
A researcher publishes more than 30 alleged N-central zero-days with working exploit code.
- Nov 9Hosted service offline
N-able takes its hosted service offline and tells on-premises customers to keep their servers offline until patches arrive.
- Nov 10Hotfix 2 ships
Build 2026.4.1.26 fixes seven CVEs.
- Nov 124EverAble2Shell released
A new chain works against servers running Hotfix 2 when the Legacy API is enabled.
- +5 hrsServers compromised
About five hours later, attackers are compromising servers. Researchers count at least 21, 19 of them on Hotfix 2.
- NextAnother hotfix coming
N-able says another hotfix is coming.
On November 8, a researcher published more than 30 alleged zero-days for N-central with working exploit code, saying N-able had failed to patch flaws he reported earlier. N-able took its hosted service offline on November 9 and told on-premises customers to keep their servers offline until patches arrived.
N-able shipped Hotfix 2 (build 2026.4.1.26) on November 10, fixing seven CVEs. Two days later, the researcher released a new chain, 4EverAble2Shell, that works against servers running that hotfix when the Legacy API is enabled. Attackers were compromising servers about five hours later.
Researchers counted at least 21 compromised servers, 19 of them on Hotfix 2, with attackers deploying cryptocurrency miners. A companion release, Agent2Shell, targets N-central’s Windows agents and, according to the researcher, gives SYSTEM access on managed devices. N-able says another hotfix is coming.
This is not N-central’s first crisis this year. In August, attackers used an authentication bypass to gain admin access and then reached managed systems through the platform’s Take Control feature. In September, N-able patched a maximum-severity flaw that it later confirmed had been exploited against a handful of customers.
Attackers have gone after RMM platforms before, including Kaseya VSA and ConnectWise ScreenConnect, because one compromised server reaches every customer it manages.
Insights & Recommendations
- Ask your MSP. Find out whether your provider uses N-central, which build it runs, whether the Legacy API is disabled, and what it found when it checked for compromise.
- Disable the Legacy API and get the console off the internet. N-able says 4EverAble2Shell requires the Legacy API, which is off by default. Expose the N-central console only through a VPN or an IP allowlist.
- Patch, then keep patching. Install Hotfix 2, upgrade every agent and probe, and apply the next hotfix as soon as N-able releases it.
- Hunt for compromise. Check N-able’s published indicators of compromise, look for unexpected admin accounts and crypto-mining processes, and review commands that N-able components ran on managed Windows endpoints.
- Protect RMM like a domain controller. Limit who can sign in, require phishing-resistant MFA, and monitor console activity.
How nGuard Helps
- Risk Assessments evaluate third-party risk, including the remote management tools your MSP uses inside your environment.
- External Penetration Testing finds remote management consoles and admin interfaces exposed to the internet before attackers do.
- If an N-central server or managed endpoint may be compromised, Cybersecurity Incident Response helps scope the intrusion, contain it, and support recovery.
Outlook Zero-Day Leaks Windows Sign-In Hashes From One Opened File
The only exploited flaw in Microsoft’s November Patch Tuesday sits in Outlook, and it hands attackers a user’s Windows sign-in hash. Microsoft fixed more than 1,000 vulnerabilities this month, its second straight month above that mark and well past the then-record 974 we covered in September.
CVE-2026-124194 is triggered by a crafted message file. When a user opens it, Outlook connects to an attacker-controlled server without the usual external-content warning, and Windows sends that server the user’s NTLMv2 hash. Attackers can relay the hash to sign in as the user or crack it to recover the password.
How one opened file leaks a sign-in hash
It is the same idea behind the 2023 Outlook zero-day we covered, which also leaked NTLM hashes to attacker servers. Microsoft rates this one only Moderate, and the fix ships through the Office update, not Windows Update, so a Windows patch cycle alone will miss it. CISA gave federal agencies until November 24.
Still unpatched: SignDrop
Patch Tuesday also brought a new unpatched exploit. Nightmare Eclipse, the researcher who has published a string of Windows zero-days since April, released SignDrop the same day. It turns a standard user into SYSTEM through the Windows Common Log File System, the November updates do not fix it, and Microsoft has not responded.
Insights & Recommendations
- Push the Office update, not just Windows. Confirm every device running classic Outlook has the November 2026 Office security update, including Microsoft 365 Apps on every update channel.
- Block outbound SMB. Block TCP 445 from workstations and servers to the internet, at the perimeter and on laptop host firewalls, so a leaked connection never leaves the network.
- Shrink NTLM. Audit outgoing NTLM, restrict it where applications allow, and add privileged accounts to the Protected Users group.
- Close relay paths. Require SMB signing, LDAP signing and channel binding, and Extended Protection on AD CS and Exchange, so a stolen hash cannot be replayed.
- Plan for SignDrop. With no patch available, watch endpoints for unexpected privilege escalation and assume a phished standard user can become SYSTEM.
How nGuard Helps
- Internal Penetration Testing shows whether a captured NTLM hash can be relayed or cracked into wider access, and which settings would stop it.
- Vulnerability Management verifies that Office updates, not just Windows updates, actually reached every endpoint.
- Social Engineering assessments test whether employees open unexpected attachments and message files.
Critical Apache HTTP/2 Flaw Found by Anthropic and Google
Apache patched a critical flaw in its HTTP Server on November 11 that could let an
unauthenticated attacker run code on any affected server with HTTP/2 enabled. The bug
is a race condition in mod_http2, the module that handles HTTP/2 traffic. Red Hat scores it 9.8 and says an attacker could exploit it remotely to run commands and take control of the
server process.
Only servers with HTTP/2 enabled are affected, but that is common on HTTPS front ends,
and many distribution and container images ship mod_http2 ready to turn on.
Servers that moved to versions 2.4.67 through 2.4.70 to fix earlier mod_http2 bugs are still vulnerable to this one.
Apache HTTP Server versions
Affected only with HTTP/2 enabledCloud environments Wiz sees
These counts do not check whether HTTP/2 is on.
Wiz found affected Apache versions in 64% of the cloud environments it sees, and publicly exposed instances in 51%, though those counts do not check whether HTTP/2 is on. No exploitation had been reported as of November 13, and the flaw is not in CISA’s KEV catalog. Still, Red Hat says public research has demonstrated a reliable exploit, and public proof-of-concept code already crashes servers.
Anthropic’s OSS Scanner reported the flaw on October 26, and Google Cloud researchers found it independently on November 3. Researchers are calling it an AI-discovered vulnerability, the trend we flagged in our Claude Mythos coverage. And as with PHP2Shell, updating the application on top does nothing: WordPress security updates cannot fix a flaw in the web server underneath.
From AI discovery to patch
- Oct 26 Anthropic’s OSS Scanner reports the flaw
- Nov 3 Google Cloud researchers find it independently
- Nov 11 Apache ships the fix in 2.4.71
- Nov 13 No exploitation reported, and not in CISA’s KEV catalog
Insights & Recommendations
- Upgrade to 2.4.71. Install Apache HTTP Server 2.4.71 or your vendor’s patched packages. On Red Hat Enterprise Linux 9 and 10, apply Red Hat’s updated
mod_http2packages. - Disable HTTP/2 if you cannot patch yet. Removing
h2andh2cfrom theProtocolsdirective temporarily closes the attack path. - Find the Apache you did not know you had. Check reverse proxies, load balancers, appliance management interfaces, and container images, not just your main web servers.
- Do not wait for exploitation. A reliable exploit has already been demonstrated in research, and the patch has been public since November 11.
How nGuard Helps
- Vulnerability Management identifies affected Apache versions across your environment and tracks them to remediation.
- External Penetration Testing confirms which internet-facing servers speak HTTP/2 and can actually be reached.
- Configuration Security Audits review web server and reverse proxy settings, including which protocols and modules are enabled.
AI Coding Agents Are Leaking Secrets to Public GitHub Repositories
An AI coding assistant’s convenience setting is putting real passwords and keys into
public GitHub repositories. Security firm Apiiro found 2,539 secrets in 1,994 public
repositories, in files that AI coding agents wrote and developers committed. Most sat
in .claude/settings.local.json, the file Claude Code uses to remember
commands a developer approved.
When a developer picks “Yes, and don’t ask again,” Claude Code can save the exact command,
including any password or token typed into it, in plain text. Claude Code keeps that
file out of git through the user’s own global settings, not the project’s .gitignore, so the protection does not travel with the repository.
How a secret ends up on public GitHub
The exposed secrets include 379 database passwords, 43 SSH logins for public servers, 42 GitHub, GitLab, and npm tokens, and 23 AWS keys. One company the researchers notified confirmed its key was live. Standard scanners missed many of them: of 1,122 files holding a secret, gitleaks flagged 564, TruffleHog flagged 643, and neither caught 215.
The exposed secrets include
Of 1,122 files holding a secret
Anthropic’s head of Claude Code said the company is looking into how to prevent the exposure. The risk is not limited to one tool, though: Apiiro also found secrets in agent chat transcripts, memory files, and Copilot auto-approve settings that developers committed.
Exposed secrets do not sit unnoticed. In October, the GhostAction campaign pushed 72,000 commits to 23,000 public repositories, and its newer payloads scan for AWS keys, GitHub tokens, and AI service secrets. AI agents are adding to a pile we already covered in October, when researchers found more than 543,000 active credentials exposed on GitHub.
Insights & Recommendations
- Search for agent files, including history. Look across your GitHub organization for
.claude/settings.local.jsonand.claude/settings.json, and check git history and forks even where the files are gone. - Rotate first, clean up second. Change every password, token, and key those files have ever held, then remove the files.
- Keep agent folders out of source control. Add agent settings files to each repository’s
.gitignoreand your organization’s templates, and treat folders like.claude/,.cursor/, and.gemini/the way you treat.envfiles. - Keep secrets out of commands. Reference secrets as variables from a secrets manager and approve command patterns instead of single commands.
- Tune your scanners. Make sure pre-commit and CI scanning cover git history and catch passwords typed inside commands, not just known key formats.
How nGuard Helps
- An AI Usage & Risk Assessment identifies which AI tools your teams use, how they are configured, and where they touch sensitive data.
- An AI Strategic Security Assessment reviews governance, policy, and secure development practices for AI-assisted work.
- AI Penetration Testing tests AI tools and integrations for the ways they can expose data.
Wrap
This week’s stories share one lesson: a fix is where the work starts, not where it ends. N-central servers on the latest hotfix were compromised within hours, the Outlook fix ships in the Office channel that many patch cycles skip, Apache’s flaw sits beneath every application running on it, and leaked secrets stay usable long after the file that held them is deleted.
Verify that fixes actually landed, keep management interfaces off the internet, rotate any credential that may have been exposed, and know which third parties hold the keys to your environment. The organizations that come through weeks like this are the ones that check, not the ones that assume.
Know where your attack paths lead? nGuard helps organizations find exposed management consoles, verify patches across every endpoint, test for credential relay and leaked secrets, and respond when something gets through.
Talk to nGuard
