Try reloading the player or open this episode directly on YouTube.
🎂 Ship It Weekly turns one
Our one-year episode
Episode 1 went out on November 20, 2025, and this one went out a year later to the day. Thank you for listening, for sharing episodes, and for sending stories our way.
This episode of Ship It Weekly covers critical security issues, including the Java4Shell flaw in OpenJDK, a hijacked Microsoft agent-framework package on PyPI, and an exploited Confluence vulnerability.
Now Playing
Java4Shell OpenJDK Flaw, Microsoft Agent Framework PyPI Package Hijacked, Confluence Axis RCE Exploited & FAA Ground Stop Traced to a Router
Ship It Weekly
0:00Current chapter: Cold open18:25
Chapters
Jump to a section in this episode.
Speed & share
Transcript
A memory bug in Java's Base64 decoder, nicknamed
Java4Shell got patched Tuesday, and your CPU
decides whether you are exposed. Microsoft's
Agent Framework package on PyPI shipped a credential
stealer that ran every time Python started. And
attackers went from an old Confluence SOAP library
to Microsoft 365 in about 12 hours. I'm Brian
Teller from Teller's Tech, and this is Ship It
Weekly.
Welcome back to Ship It Weekly, the show
about the DevOps, SRE, cloud, platform, and security
stories that matter when you are the person keeping
the thing running at three in the morning. For
the weekly story list and source links, check
out oncallbrief.com. For past episodes and show
notes, head over to shipitweekly.fm. This week,
the Java4Shell OpenJDK flaw, Microsoft's hijacked
agent framework package, a Confluence bug under
active attack, and the router change behind the
FAA's ground stop. Then we have a quick lightning
round, and a human closer about why your root
cause analysis might not be working. Let's get
into it.
First up, Java4Shell, CVE-2026-114938,
a memory corruption bug in the OpenJDK HotSpot
JVM. Oracle and the OpenJDK project fixed it
on Tuesday, November 17th. It affects Java 17
and newer, including Corretto, Zulu, and other
downstream builds. Calif, which found it, rates
it 9.2 critical. Oracle says 8.1 because it's
hard to exploit. The mechanism is narrow. To
decode Base64 quickly, the JVM swaps the normal
Java code for handwritten assembly that uses
AVX-512 instructions. That assembly miscounts
the output size when the input ends with an unexpected
number of equal signs. The length math underflows,
and the JVM writes past the end of a buffer on
the heap. Calif says 64 letter A's followed
by one equal sign is enough to trigger it on
newer JDKs. The vulnerable path only runs on
x86 CPUs with the AVX-512 VBMI instructions. Intel
Xeons from Ice Lake on and AMD Zen 4 and newer.
ARM64 isn't affected, so your Graviton nodes
aren't either. On Linux, look for the AVX-512 VBMI
flag in /proc/cpuinfo. VulnCheck says crashing
a service is easy. Code execution is much harder.
It needs heap grooming plus a separate memory
leak. Calif says it got code execution in apps
like Jira and Jenkins, and it's holding those
exploits back. As of Wednesday, GreyNoise had
reported exploit attempts, but no successful
break-ins. So patch the JDK on your normal schedule,
but find all of it. Base images, runtimes, bundled
in vendor appliances, that old build agent. If
a host can't be patched this week, Wiz says
a JVM flag that turns off the Base64 intrinsics
closes the vulnerable path, at some cost in performance.
The bug is in Java, but the inventory problem
is yours.
Next, Microsoft's Agent Framework package
on PyPI shipped a credential stealer. On Tuesday
evening, UTC, two malicious versions, 1.26.1
and 1.26.2 went up 33 minutes apart. Both were
off PyPI within an hour of the first one. The
package gets about 190,000 downloads a month,
and the last clean version is 1.26.0. How the
attacker got publishing access hasn't been disclosed.
What makes this one worth understanding is how
it ran. Neither version needed you to import
anything. Each one dropped a .pth file into site
packages. Python reads those files every time
the interpreter starts. And it runs any line
in them that begins with the word import. So
the stealer ran with your tests, your app, any
script in that environment, even pip. It collected
environment variables with names like token,
secret, and key, plus credential files for the
Azure CLI, npm, Docker, and Git. Then it asked
Azure for fresh managed identity tokens for Key
Vault, Storage, and Resource Manager, plus Azure
DevOps OIDC tokens. The second version added
GitHub Actions OIDC tokens in workflows allowed
to request one. All of it went out in a single
post to a webhook.site address. Those fresh
tokens change the cleanup. Rotating your stored
secrets doesn't touch a token the attacker requested
as you. Wiz says those tokens last up to 24 hours.
So check what your managed identities and federated
credentials actually did. Search your lock files,
CI logs, and images for 1.26.1 and 1.26.2
and pin 1.26.0. Look for webhook.site in your
proxy and DNS logs. If it ran, rotate everything
that environment could reach and review sign-ins
for every identity it could use.
Third, attackers are exploiting Confluence Data Center
through CVE-2026-90436, a 9.8 critical bug in
Apache Axis 1. Axis is the old SOAP engine behind
Confluence's remote API. The flaw is unsafe reflection.
A request gets to choose which Java class is
loaded, and no login is needed. Axis 1 is end
of life, and Apache isn't planning a fixed release.
Rapid7 saw it used against a customer last Friday,
November 13th. Unauthenticated requests hit the
SOAP endpoint, and within seconds, the Confluence
Java process was starting bash and curl. About
three minutes in, the attacker logged into MySQL
with Confluence's own database credentials, read
pages, and searched attachments for credentials,
tokens, VPN configs, and private keys. About
four hours in, they were querying a production
data warehouse with a service account. About
12 hours in, they signed into Microsoft 365 as
an integration service account using legacy authentication.
I think the interesting part is the setting that
makes you vulnerable. The remote API has been
off by default for new installs since Confluence
3.0. So the exposed servers are the ones where
someone switched it on for an integration, maybe
a year ago, and nobody switched it off. If you
run Confluence Data Center, go to General Configuration,
then Further Configuration, and check Remote API.
Atlassian fixed it in 9.2.28, and 10.x doesn't
have the feature. Turning the setting off works
without a restart, but anything that integrates
through it will break. Atlassian Cloud isn't
affected. N-able also says attackers on N-central
strongly correlate with this flaw, and it's fixed
in 2026.4 Hotfix 3. A wiki is only as safe as
the credentials people have pasted into it.
Fourth, the FAA's nationwide ground stop last Friday.
At about 3:30 p.m. Eastern on November 13th,
the FAA halted flights across the country. Traffic
started moving again at 8:15 that evening. FedScoop
says that's only the second time since September
11th. Operators canceled about 12,000 flights
that day, not all of them because of the stop.
The FAA says there's no sign of a cyber attack.
The preliminary story is a network change. The
FAA's early findings point to two sites, one
of them a radio site in El Dorado, Arkansas,
where its network provider L3Harris was installing
a new router as part of a move from copper to
fiber. Officials say that router flooded the
network with about a thousand connection attempts
every four to five seconds. And the legacy TDM
equipment behind it went unstable. It hit 17
of the 22 centers that handle high altitude traffic.
What I like about this story is what came next.
FAA administrator Brian Bedford said they spent
about 45 minutes trying to find the cause. That
maintenance is logged by hand, so they were going
through sheets of paper to find out what had
changed. L3Harris blocked both sites, and the
network settled. The FAA has since disabled more
than 60 other Juniper routers as a precaution,
and says it will rethink when this kind of work
gets done. This one happened midday. The questions
carry over. When something breaks, can you list
every change from the last hour? Can you search
that list? Or is it someone's notes? And does
your maintenance window line up with your quietest
traffic? Or with when the contractor is available?
Quick lightning round. First, wfx-loot. Starting
November 10th, attackers used stolen GitHub credentials
from 4,837 accounts to probe 61,764 repositories
for write access, according to Aikido. Where they
had it, they pushed a credential stealer into
a workflow or an npm install hook, then reverted
it and deleted the run within about a minute.
Look for a branch called wfx-docs or a gap in
your workflow run numbers. And rotating secrets
isn't enough. Where it could, it also rewrote
AWS role trust policies to accept any account
and added Azure federated credentials. Second,
PHP is back. A month after PHP2Shell, PHP
shipped another emergency release on Thursday,
with seven critical fixes all rated 9.2. At
least two of them get around October's PHP2Shell
patches, so being patched last month isn't
enough. One, nicknamed phpAfterShell, was found
by depthfirst with help from its AI models. An
interrupted session write can leave a file that's
half old, half new, and PHP trusts it anyway.
Update to 8.5.13, 8.4.28, 8.3.37, or 8.2.36.
If you can't yet, turn off session upload
progress. And if you are still on 8.1 or older,
there's no official fix. And third, Kubernetes
now has a date for removing kube-proxy's IPVS
mode: version 1.40. Back in August, we talked
about it starting to retire. On 1.36 or later,
with a 5.13 or newer kernel, SIG Network recommends
nftables, which is a config change and a kube-proxy
restart.
The human closer this week comes
from Oakley Hall in a piece featured in SRE Weekly
called Why Your RCAs Aren't Working. His point
is that you can write a lot of root cause analysis
documents without getting better at anything.
He gives four reasons, and two of them fit this
week really well. The first is that we collect
documents, not data. There's no table that you
can query across incidents, by change type, by
platform, or by the hour a change went out. So
spotting a pattern depends on whoever has read
most of them and how much they remember. The
second is that systematic problems get assigned
to whichever team happened to hit them. That
team can fix its own service. It usually can't
change a shared platform, an unsafe default,
or a dependency everybody uses. So it builds
a local workaround. The RCA gets closed and the
next team finds the same problem. This week fits
both. The FAA spent 45 minutes finding a router
change, partly because the maintenance record
was on paper. That's documents instead of data,
about as literal as it gets. And Java4Shell
isn't a ticket for every app team. It's work
for whoever owns the base images and the JDK
fleet. Same with an end-of-life SOAP library
inside Confluence, or a .pth file that runs before
your code does. Neither belongs to the team that
noticed it first. His fix is a formal way to
escalate those actions into funded platform work,
with an owner who has the authority to finish
them. Otherwise, you get a name in the document
and no path to a fix. That's it for this week
of Ship It Weekly. Follow or subscribe wherever
you are watching or listening. You can find the
weekly story list and source links at oncallbrief.com
and past episodes and show notes at shipitweekly.fm.
Quick thing before I let you go. Today's
November 20th, and Ship It Weekly is officially
one year old. Episode one came out a year ago
today, and this episode is episode 86. I started
this because keeping up with DevOps is hard,
and I wanted something practitioners could actually
use. In year one, this show won a Davey Award
gold, a dotCOMM gold, and a Viddy. And we're a DevOps
Dozen finalist in two categories. The show for
Best DevOps Community Program, and me for Enterprise
DevOps Champion of the Year. If this show's helped
This week's stories all lived one layer below where most of us look.
Not in the application. Not in the code anyone on your team wrote last sprint. Underneath it: an instruction set on the CPU, a file Python reads before your program starts, a SOAP library older than most of the people running it, and a router at a remote radio site.
That's the layer we inherit. We rarely inventory it, and we almost never own it on purpose.
Start with Java4Shell.
CVE-2026-114938 is a memory corruption bug in the OpenJDK HotSpot JVM, fixed on Tuesday, November 17th. It affects Java 17 and newer, and every downstream build: Corretto, Temurin, Zulu and the rest. To decode Base64 quickly, the JVM swaps the normal Java code for hand-written assembly that uses AVX-512 instructions. That assembly miscounts the output size when the input ends with an unexpected number of equals signs, and the JVM writes past the end of a heap buffer. Calif, which found it, scores it 9.2. Oracle says 8.1, because getting code execution out of it is hard.
What I find interesting is how narrow the path is. It only runs on x86 CPUs with the AVX-512 VBMI instructions: Intel Xeons from Ice Lake on, and AMD Zen 4 and newer. ARM64 never takes it, so Graviton nodes are fine. Two hosts running the exact same JDK build can have different answers to "am I exposed?", depending on the silicon underneath them.
My take: that's the part an inventory usually misses. We track JDK versions, sometimes. We almost never track which CPU family each JVM landed on. And the JDKs that get missed aren't in the service you deploy every day. They're in base images, in a runtime bundled inside a vendor appliance, and on that build agent nobody has rebuilt since spring.
So patch on your normal schedule, but go find all of it. On Linux, grep avx512vbmi /proc/cpuinfo tells you whether a host can take the vulnerable path. If one can't be patched this week, Wiz says -XX:-UseBASE64Intrinsics closes the path at some cost in performance. As of Wednesday, GreyNoise had seen exploit attempts but no successful break-ins.
The bug is in Java. The inventory problem is yours.
Then Microsoft's agent-framework package on PyPI.
On the evening of November 17th, UTC, two malicious versions, 1.26.1 and 1.26.2, went up 33 minutes apart. Both were gone within an hour of the first. The package gets about 190,000 downloads a month, and the last clean version is 1.26.0. How the attacker got publishing access hasn't been disclosed.
The mechanism is the reason I put it second. Neither version needed you to import anything. Each dropped a .pth file into site-packages. Python reads those files every time the interpreter starts, and it runs any line in them that begins with import. So the stealer ran with your tests, your app, any script in that environment, even pip itself.
It also didn't stop at what was stored. It collected environment variables and credential files, and then it asked Azure for fresh managed identity tokens, and for Azure DevOps OIDC tokens. The second version added GitHub Actions OIDC. Those tokens were requested as you. Rotating your stored secrets doesn't touch them. Wiz says they last up to 24 hours.
That changes what cleanup means. Search lockfiles, CI logs and images for 1.26.1 and 1.26.2, and pin 1.26.0. Look for webhook.site in proxy and DNS logs. And if it ran, rotate what the environment could reach, then go read what your managed identities and federated credentials actually did that day.
Then Confluence.
CVE-2026-90436 is a 9.8 critical bug in Apache Axis 1, the old SOAP engine behind Confluence's remote API. A request gets to choose which Java class is loaded, and no login is needed. Axis 1 is end of life, and Apache isn't planning a fix.
Rapid7 saw it used against a customer on November 13th. Within seconds, the Confluence Java process was starting bash and curl. About three minutes in, the attacker was in MySQL with Confluence's own database credentials, searching pages and attachments for credentials, tokens, VPN configs and private keys. About four hours in, a production data warehouse. About twelve hours in, Microsoft 365, through legacy authentication.
The setting is the story. The remote API has been off by default for new installs since Confluence 3.0. The exposed servers are the ones where someone turned it on for an integration, maybe years ago, and nobody turned it off. Check General Configuration, then Further Configuration, then Remote API. Atlassian fixed it in 9.2.28, and 10.x doesn't have the feature. Cloud isn't affected. N-able says attacks on N-central strongly correlate with this flaw, and its fix is 2026.4 Hotfix 3.
A wiki is only as safe as the credentials people have pasted into it. My guess is most of us have never searched ours.
Then the FAA.
At about 3:30 PM Eastern on Friday, November 13th, the FAA stopped flights nationwide until 8:15 that evening. There's no sign of a cyberattack. The preliminary story is a network change: at a radio site in El Dorado, Arkansas, L3Harris was installing a new router as part of a copper-to-fiber move. Officials say it flooded the network with about a thousand connection attempts every four to five seconds, and the legacy equipment behind it went unstable. Seventeen of the 22 high-altitude centers were affected.
The part that stuck with me: the FAA Administrator said they spent about 45 minutes finding the cause, partly because that maintenance was logged by hand. They went through sheets of paper to find out what had changed. The FAA has since disabled more than 60 other Juniper routers, and says it will rethink when this kind of work gets done. This one happened midday.
So ask the plain questions. When something breaks, can you list every change from the last hour? Can you search that list, or is it somebody's notes? Does your maintenance window match your quietest traffic, or the contractor's calendar?
The human closer is where this week comes together for me.
Oakley Hall's piece, "Why Your RCAs Aren't Working," gives four reasons, and two of them fit. The first: we collect documents, not data. There's no table you can query across incidents by change type, platform or hour, so spotting a pattern depends on who has read the most RCAs. The FAA's paper log is that, about as literally as it gets.
The second: systemic problems get assigned to whichever team hit them first. That team can fix its own service. It usually can't change a shared base image, a JDK fleet, an end-of-life library inside Confluence, or how a package installs. So it builds a workaround, the RCA closes, and the next team finds the same problem.
Every story this week was work for whoever owns the layer underneath. That only works if someone actually does.
A corrective action needs an owner who can actually finish it.
One last thing: thank you for listening, for sharing episodes, and for sending stories my way. If the show's helped you this year, the DevOps Dozen vote is open through December 31st.
This week on Ship It Weekly, Java4Shell: a memory bug in the OpenJDK HotSpot JVM's Base64 decoder, fixed Tuesday, that only bites on x86 CPUs with AVX-512 VBMI. Microsoft's agent-framework package on PyPI shipped a credential stealer in two versions that were live for under an hour. Attackers are exploiting Confluence Data Center through CVE-2026-90436, a bug in its old Apache Axis SOAP library, and in one case reached Microsoft 365 in about twelve hours. And the FAA's preliminary cause for last Friday's nationwide ground stop: a new router at a remote site that flooded its network.
The thread: the layer under the code. A CPU instruction set, a .pth file that runs before your code does, a 20-year-old SOAP library, and a router swap logged on paper.
In the lightning round: wfx-loot, the GitHub campaign that probed 61,764 repos; PHP's second emergency release in a month; and kube-proxy's IPVS mode leaving Kubernetes in 1.40.
The human closer: Oakley Hall on why your RCAs aren't working. Documents without data, and systemic fixes handed to whichever team hit them first.
Ship It Weekly turns one today. Thank you for listening. The show is a 2026 DevOps Dozen finalist for Best DevOps Community Program or Event, and I'm one for Enterprise DevOps Champion of the Year. Voting is open through December 31.
Highlights
Check /proc/cpuinfo for avx512vbmi, and the JVM flag that closes the path
Pin agent-framework 1.26.0 and search for 1.26.1 and 1.26.2
Is Confluence's Remote API on? Where to check, and the fixed version
How a router swap in Arkansas stopped flights across the country
A corrective action needs an owner who can finish it
25 years in production: DevOps, SRE, platform, and cloud. Host of Kube Signals on KubeFM. Grafana Champion; Confluent Community Catalyst; IBM Influencer 2026; DevOps Institute & ITIL Ambassador.
The Layer Under the Code
This week's stories all lived one layer below where most of us look.
Not in the application. Not in the code anyone on your team wrote last sprint. Underneath it: an instruction set on the CPU, a file Python reads before your program starts, a SOAP library older than most of the people running it, and a router at a remote radio site.
That's the layer we inherit. We rarely inventory it, and we almost never own it on purpose.
Start with Java4Shell.
CVE-2026-114938 is a memory corruption bug in the OpenJDK HotSpot JVM, fixed on Tuesday, November 17th. It affects Java 17 and newer, and every downstream build: Corretto, Temurin, Zulu and the rest. To decode Base64 quickly, the JVM swaps the normal Java code for hand-written assembly that uses AVX-512 instructions. That assembly miscounts the output size when the input ends with an unexpected number of equals signs, and the JVM writes past the end of a heap buffer. Calif, which found it, scores it 9.2. Oracle says 8.1, because getting code execution out of it is hard.
What I find interesting is how narrow the path is. It only runs on x86 CPUs with the AVX-512 VBMI instructions: Intel Xeons from Ice Lake on, and AMD Zen 4 and newer. ARM64 never takes it, so Graviton nodes are fine. Two hosts running the exact same JDK build can have different answers to "am I exposed?", depending on the silicon underneath them.
My take: that's the part an inventory usually misses. We track JDK versions, sometimes. We almost never track which CPU family each JVM landed on. And the JDKs that get missed aren't in the service you deploy every day. They're in base images, in a runtime bundled inside a vendor appliance, and on that build agent nobody has rebuilt since spring.
So patch on your normal schedule, but go find all of it. On Linux,
grep avx512vbmi /proc/cpuinfotells you whether a host can take the vulnerable path. If one can't be patched this week, Wiz says-XX:-UseBASE64Intrinsicscloses the path at some cost in performance. As of Wednesday, GreyNoise had seen exploit attempts but no successful break-ins.The bug is in Java. The inventory problem is yours.
Then Microsoft's agent-framework package on PyPI.
On the evening of November 17th, UTC, two malicious versions, 1.26.1 and 1.26.2, went up 33 minutes apart. Both were gone within an hour of the first. The package gets about 190,000 downloads a month, and the last clean version is 1.26.0. How the attacker got publishing access hasn't been disclosed.
The mechanism is the reason I put it second. Neither version needed you to import anything. Each dropped a
.pthfile into site-packages. Python reads those files every time the interpreter starts, and it runs any line in them that begins withimport. So the stealer ran with your tests, your app, any script in that environment, even pip itself.It also didn't stop at what was stored. It collected environment variables and credential files, and then it asked Azure for fresh managed identity tokens, and for Azure DevOps OIDC tokens. The second version added GitHub Actions OIDC. Those tokens were requested as you. Rotating your stored secrets doesn't touch them. Wiz says they last up to 24 hours.
That changes what cleanup means. Search lockfiles, CI logs and images for 1.26.1 and 1.26.2, and pin 1.26.0. Look for
webhook.sitein proxy and DNS logs. And if it ran, rotate what the environment could reach, then go read what your managed identities and federated credentials actually did that day.Then Confluence.
CVE-2026-90436 is a 9.8 critical bug in Apache Axis 1, the old SOAP engine behind Confluence's remote API. A request gets to choose which Java class is loaded, and no login is needed. Axis 1 is end of life, and Apache isn't planning a fix.
Rapid7 saw it used against a customer on November 13th. Within seconds, the Confluence Java process was starting bash and curl. About three minutes in, the attacker was in MySQL with Confluence's own database credentials, searching pages and attachments for credentials, tokens, VPN configs and private keys. About four hours in, a production data warehouse. About twelve hours in, Microsoft 365, through legacy authentication.
The setting is the story. The remote API has been off by default for new installs since Confluence 3.0. The exposed servers are the ones where someone turned it on for an integration, maybe years ago, and nobody turned it off. Check General Configuration, then Further Configuration, then Remote API. Atlassian fixed it in 9.2.28, and 10.x doesn't have the feature. Cloud isn't affected. N-able says attacks on N-central strongly correlate with this flaw, and its fix is 2026.4 Hotfix 3.
A wiki is only as safe as the credentials people have pasted into it. My guess is most of us have never searched ours.
Then the FAA.
At about 3:30 PM Eastern on Friday, November 13th, the FAA stopped flights nationwide until 8:15 that evening. There's no sign of a cyberattack. The preliminary story is a network change: at a radio site in El Dorado, Arkansas, L3Harris was installing a new router as part of a copper-to-fiber move. Officials say it flooded the network with about a thousand connection attempts every four to five seconds, and the legacy equipment behind it went unstable. Seventeen of the 22 high-altitude centers were affected.
The part that stuck with me: the FAA Administrator said they spent about 45 minutes finding the cause, partly because that maintenance was logged by hand. They went through sheets of paper to find out what had changed. The FAA has since disabled more than 60 other Juniper routers, and says it will rethink when this kind of work gets done. This one happened midday.
So ask the plain questions. When something breaks, can you list every change from the last hour? Can you search that list, or is it somebody's notes? Does your maintenance window match your quietest traffic, or the contractor's calendar?
The human closer is where this week comes together for me.
Oakley Hall's piece, "Why Your RCAs Aren't Working," gives four reasons, and two of them fit. The first: we collect documents, not data. There's no table you can query across incidents by change type, platform or hour, so spotting a pattern depends on who has read the most RCAs. The FAA's paper log is that, about as literally as it gets.
The second: systemic problems get assigned to whichever team hit them first. That team can fix its own service. It usually can't change a shared base image, a JDK fleet, an end-of-life library inside Confluence, or how a package installs. So it builds a workaround, the RCA closes, and the next team finds the same problem.
Every story this week was work for whoever owns the layer underneath. That only works if someone actually does.
A corrective action needs an owner who can actually finish it.
One last thing: thank you for listening, for sharing episodes, and for sending stories my way. If the show's helped you this year, the DevOps Dozen vote is open through December 31st.
Additional Links
Attackers compromise Microsoft's agent-framework (The Hacker News)
CVE-2026-90436 (Atlassian)
phpAfterShell (depthfirst)
The GitHub Actions secret stealer and its indicators (StepSecurity)
Java4Shell (Calif)
SRE Weekly #543