Latest / Connecting the Dots / Software Supply Chain Attacks, AI Tool Exploits, and Geopolitical Tech Friction

Transcript

Software Supply Chain Attacks, AI Tool Exploits, and Geopolitical Tech Friction · Connecting the Dots

• 0:00
  Today's tech news moves fast, but understanding
• 0:03
  it shouldn't be hard. That's why Snarfl Solutions
• 0:06
  Group brings you Connecting the Dots, where we
• 0:09
  break down what's happening in tech and why it
• 0:11
  matters to you. Now, let's dive in with your
• 0:14
  hosts. The code running your company's web app
• 0:18
  was secretly checking your passport before deciding
• 0:22
  whether to just burn your servers to the ground.
• 0:25
  Welcome to Connecting the Dots. broadcasting
• 0:27
  live from Smartful Solutions Group right here
• 0:29
  in Sacramento, California. I'm Alex. And I'm
• 0:32
  Morgan. And today is Tuesday, May 12, 2026. Before
• 0:37
  we get into today's news, let's get you caught
• 0:39
  up on the world right outside your window and
• 0:42
  the numbers driving your wallet. If you are stepping
• 0:45
  out in Sacramento today, you are in for an absolutely
• 0:48
  gorgeous one. It is sunny, sitting right at 88
• 0:51
  degrees Fahrenheit with a nice light breeze anywhere
• 0:54
  from 3 to 12 miles per hour. Oh, that is perfect
• 0:56
  weather. Just beautiful. But if you have family
• 0:59
  at a state or maybe some travel plans, definitely
• 1:01
  keep your head on a swivel. There are currently
• 1:02
  289 active US weather alerts right now. Wow,
• 1:07
  really? Yeah, 289. And 77 of those are over land.
• 1:11
  So always check weather .gov before you. you
• 1:13
  know, hit the road. Which is super important,
• 1:16
  especially as we head deeper into the spring
• 1:18
  travel, Seardon. You definitely don't want to
• 1:19
  get caught in one of those unexpected cross -country
• 1:21
  systems. Oh, absolutely not. And looking at the
• 1:24
  markets, we saw a pretty solid start to the week
• 1:26
  on Monday. The Dow Jones closed up 155 .40 points,
• 1:32
  which is a 0 .31 % bump. Nice. Yeah. And the
• 1:35
  S &P 500 kind of mirrored that exact energy,
• 1:38
  closing up 27 .53 points or 0 .37%. Yeah. Over
• 1:42
  in the crypto space, Bitcoin is currently sitting
• 1:44
  at 80 ,634 .76, so it's down just a hair, like
• 1:49
  1 .33 % for the day. Just a little dip. Yeah,
• 1:52
  nothing crazy. And for our sports fans, a quick
• 1:54
  shout out to the SF Giants, who took down the
• 1:56
  Dodgers 9 -3 yesterday. Oh man, that was a game.
• 1:59
  It was. Matt Gage was the winning pitcher just
• 2:01
  through a phenomenal game. Meanwhile, the Kings
• 2:03
  and the A's had the night off, so everyone is
• 2:06
  rested up and ready for the rest of the week.
• 2:07
  Yeah, it's good to see the Giants carrying that
• 2:09
  momentum. Hopefully the markets maintain that
• 2:11
  similar steady climb. We will definitely see.
• 2:14
  But before we get into the heavy -hitting tech
• 2:16
  news shaping your week, today's episode is brought
• 2:18
  to you by Strong Coffee, the premium instant
• 2:21
  coffee that tastes like a traditional brew, but
• 2:23
  is packed with 15 grams of protein, MCTs, and
• 2:26
  adaptogens like ashwagandha. Get clean, sustained
• 2:29
  energy without the jitters. Use promo code SNARFL
• 2:32
  for 20 % off at strongcoffeecompany .com. So
• 2:36
  let's set the stage for what is actually happening
• 2:38
  across the tech landscape today, because there
• 2:41
  is this very clear defining tension right now.
• 2:44
  We are basically looking at the invisible borders
• 2:47
  of the digital world. Yeah, because the software
• 2:49
  that your family relies on every day to bank
• 2:52
  or communicate or do your work, it's currently
• 2:54
  facing this massive, highly sophisticated attack.
• 2:57
  It's literally a war being waged on the software
• 3:00
  layer. Exactly. But all of this code, I mean,
• 3:02
  it's entirely useless without the heavy duty
• 3:04
  computing power to actually run it. And the physical
• 3:06
  hardware, you know, the actual silicon, is caught
• 3:09
  in this intense high stakes geopolitical tug
• 3:12
  of war. Right. So software wants to be borderless,
• 3:16
  freely shared across the globe. But the hardware
• 3:19
  required to run it is becoming fiercely nationalized.
• 3:22
  Which is exactly the conflict we're seeing. So
• 3:24
  let's unpack this massive supply chain attack
• 3:27
  to see exactly how that software is being targeted
• 3:29
  right now. Security researchers are tracking
• 3:32
  this new campaign and they've dubbed it. The
• 3:34
  mini shy hooloo'd worm. Like from Dune. Like
• 3:36
  from Dune, yeah. Giant sandworm. And this whole
• 3:40
  thing starts with NPM packages. For anyone listening
• 3:42
  who isn't, you know, writing code every day,
• 3:45
  think of NPM as this massive global hardware
• 3:48
  store of pre -written code. Developers use these
• 3:51
  packages as foundational bill get blocks so they
• 3:53
  don't have to just reinvent the wheel every single
• 3:56
  time they build a website. The issue is you trust
• 3:59
  that the cement you're buying from that hardware
• 4:01
  store isn't secretly dissolving your foundation.
• 4:05
  This specific attack targeted 84 of those NPM
• 4:08
  packages, primarily in a namespace called Tanstack.
• 4:12
  And Tanstack is huge, right? Oh, it's foundational.
• 4:14
  For web developers, we're talking about packages
• 4:16
  like React Router. That one alone sees over 12
• 4:20
  million downloads a week. 12 million. Week a
• 4:23
  week. Yeah, and you I pathos packages were also
• 4:27
  compromised in this Wow and the mechanics of
• 4:29
  this breach like how they actually pulled it
• 4:32
  off or just Staggering the attackers used a pattern
• 4:35
  that researchers are calling a pin request. Yeah
• 4:38
  puner quest is a I mean a malicious but brilliant
• 4:42
  piece of engineering. They didn't just brute
• 4:44
  force a password or guess an admin login. Too
• 4:46
  easy. Right. Instead, they exploited the continuous
• 4:50
  integration systems, the CI pipelines, that developers
• 4:53
  use to automatically build and test their code.
• 4:55
  Specifically, they targeted GitHub actions. The
• 4:58
  hackers essentially poisoned the cache across
• 5:00
  the trust boundary. Okay, wait, let's pause for
• 5:03
  a second. Let's break that down because that
• 5:04
  is a lot of jargon. Right. You're saying they
• 5:06
  poisoned a temporary memory cache. Right. But
• 5:09
  how does messing with temporary memory rewrite
• 5:12
  the permanent official code that is being downloaded
• 5:14
  12 million times a week? That doesn't make sense
• 5:17
  to me. Yeah, so it has to do with how automated
• 5:20
  testing works to save time. Imagine you have
• 5:22
  a secure facility, right? And a guest worker
• 5:25
  submits a prototype for review. The automated
• 5:28
  system, which is GitHub Actions here, runs a
• 5:30
  test on this new, seemingly innocent code. But
• 5:34
  during that test, the malicious code deliberately
• 5:36
  saves a compromised file into the system's shared
• 5:39
  memory, or cache. Ah, I see. Right. So later,
• 5:43
  when the legitimate, highly trusted owners of
• 5:45
  the project run their official, secure publishing
• 5:47
  process to release an update, their automated
• 5:50
  system just wants to save time. So it pulls from
• 5:53
  the cache. Exactly. It reaches into that cache,
• 5:55
  pulls out the poisoned file instead of compiling
• 5:58
  a fresh one, and inadvertently packages the malware
• 6:01
  into the official release. That is wildly clever.
• 6:04
  Yeah. And terrifying. And from what I understand,
• 6:07
  that poisoned file allowed the attackers to extract
• 6:09
  an OIDC token from the system's memory. Yes.
• 6:12
  So for context, an OpenID Connect token, an OIDC
• 6:15
  token, is basically a temporary, highly trusted
• 6:18
  digital ID badge. It's kind of like... Okay,
• 6:21
  imagine going to a really high -end restaurant
• 6:23
  and a valet takes your car. But the valet is
• 6:25
  a hacker. They steal your master key fob, the
• 6:29
  one that's programmed to open your car, your
• 6:31
  house, and your office. And they just use that
• 6:33
  fob to quietly unlock the doors and walk right
• 6:36
  in. And the most dangerous part of that valet
• 6:39
  key analogy is that the automated security system
• 6:42
  never logs the valet's actions as suspicious
• 6:45
  because it thinks that you are the one holding
• 6:47
  the key. Right, because it's your fob. Exactly.
• 6:51
  Because they possess that OIDC token, GitHub
• 6:54
  believed the hackers were the legitimate developers.
• 6:56
  And that is what allowed the attackers to publish
• 6:58
  a 2 .3 megabyte heavily obfuscated payload called
• 7:02
  routerinit .js. and obfuscated, just meaning
• 7:06
  they intentionally scrambled the code so that,
• 7:07
  you know, automated security scanners couldn't
• 7:10
  read its true intent. Right, it looks like gibberish
• 7:12
  to the scanners. But once a developer accidentally
• 7:14
  downloads this compromised package, the worm
• 7:16
  just goes to work. It actively hunts for the
• 7:19
  keys to the corporate kingdom. I mean, it scrapes
• 7:21
  the developer's environment for Amazon Web Services
• 7:24
  keys, GitHub tokens, HashiCorp Vault credentials,
• 7:27
  and even Kubernetes service accounts. But it
• 7:30
  doesn't even stop at credential theft. That's
• 7:32
  the wild part. It's a self -replicating worm.
• 7:36
  It actively works to spread itself by impersonating
• 7:38
  a legitimate AI coding assistant, the Claude
• 7:41
  code bot. Wait, it pretends to be an AI? Yes.
• 7:45
  By spoofing the email address Claude at users
• 7:48
  .norreply .github .com It just quietly commits
• 7:52
  copies of itself into other code repositories.
• 7:55
  So to anyone looking, it just makes it look like
• 7:57
  a helpful AI made a routine code adjustment.
• 8:00
  Oh, wow. That is sneaky. And the persistence
• 8:03
  mechanism is what really gets me here. Like it
• 8:05
  doesn't just steal your data and leave. It's
• 8:07
  like finding out the burglar didn't just take
• 8:09
  your valuables. They secretly hardwired a back
• 8:12
  door into your home's security panel. The worm
• 8:15
  buries itself in hidden local folders on the
• 8:17
  developer's machine so that even if the computer's
• 8:20
  fully restarted, you reboot it entirely, the
• 8:22
  malware just boots right back up. It's super
• 8:24
  persistent. Not only that, but how it communicates
• 8:26
  back to the hackers is incredibly stealthy. It
• 8:28
  exfiltrates the stolen credentials through the
• 8:31
  Session P2P network. Which is an encrypted messaging
• 8:33
  app, right? Yeah. Exactly. Session is a decentralized
• 8:36
  encrypted messaging network. By routing the stolen
• 8:39
  data through Session, the malicious traffic blends
• 8:42
  in perfectly. So to a security analyst who might
• 8:46
  be monitoring the network, it just looks like
• 8:48
  routine telemetry. Just normal data. Just normal
• 8:51
  operational data pinging back and forth from
• 8:54
  a secure chat app. It leaves almost zero footprint.
• 8:57
  Wow. So when we strip away all the tech jargon,
• 9:00
  the reality is that the foundational building
• 9:02
  blocks used to create the apps you and your family
• 9:05
  use every single day were briefly laced with
• 9:08
  a highly evasive, self -replicating credentials
• 9:10
  stealer. Which is massive on its own. But this
• 9:13
  exact same infrastructure has evolved and the
• 9:16
  consequences have shifted from, you know, silent
• 9:18
  data theft to outright targeted destruction.
• 9:21
  OK, let's look closer at how this evolved. because
• 9:23
  the Mini Shy Hulu campaign escalated straight
• 9:26
  into the booming AI ecosystem. And it just proves
• 9:29
  that open source code is no longer neutral territory.
• 9:32
  Not at all. Microsoft Threat Intelligence actually
• 9:34
  identified that a Python package for Mistral
• 9:36
  AI was also compromised, specifically version
• 9:38
  2 .4 .6. And Mistral is huge. Yeah, you already
• 9:42
  know Mistral is foundational infrastructure for
• 9:44
  enterprise AI right now. Hitting their library
• 9:46
  isn't just a random strike. it is targeting the
• 9:49
  absolute center of gravity for corporate machine
• 9:51
  learning. Right. So an AI developer imports this
• 9:54
  package, assuming it's just the standard secure
• 9:57
  Mistral AI tool. But the moment it is imported
• 10:00
  onto a Linux machine, the injected code silently
• 10:03
  reaches out to a remote IP address and downloads
• 10:06
  a secondary payload. And they deceptively named
• 10:09
  this payload transformers .pyz which is brilliant
• 10:13
  camouflage honestly hugging face which is another
• 10:16
  titan in the ai space maintains a wildly popular
• 10:18
  library called transformers oh i see so it's
• 10:21
  a disguise right if a security operations center
• 10:24
  spots a file called transformers running on an
• 10:26
  ai engineer's workstation they likely won't flag
• 10:29
  it It blends right into the expected environment
• 10:31
  of a machine learning workload. Nobody would
• 10:33
  look twice at it. But the payload itself isn't
• 10:36
  looking for EWS keys anymore. This new one has
• 10:38
  highly specific geo -fencing logic built into
• 10:41
  its core. Yeah. And to understand geofencing
• 10:44
  in this context, we have to ask how a piece of
• 10:46
  code actually knows what country it's currently
• 10:49
  sitting in. How does it do that? Well, the moment
• 10:51
  it activates, it queries the computer system
• 10:54
  clock to check the local time zone. Then it reviews
• 10:57
  the default language settings of the operating
• 10:59
  system. And finally, it cross -references local
• 11:02
  IP registries to basically build a complete geographic
• 11:05
  profile of the machine. And based on that geographic
• 11:08
  profile, it makes decisions. Like first, If it
• 11:12
  detects the system is running in Russian, it
• 11:15
  deliberately shuts down and does absolutely nothing.
• 11:17
  Which, by the way, is a classic obfuscation tactic
• 11:19
  in cybersecurity. It doesn't definitively prove
• 11:22
  the hackers are Russian. Oh, really? Yeah, often
• 11:24
  threat actors will write code to deliberately
• 11:27
  bypass certain regions to avoid drawing the ire
• 11:30
  of local authorities, you know, or specifically
• 11:32
  to throw security researchers off their trail
• 11:34
  by framing a particular nation -state. Okay,
• 11:37
  that makes sense. A little misdirection. But
• 11:39
  the next check is where the weaponization actually
• 11:41
  happens. If the payload determines the computer
• 11:44
  is operating in Israel or Iran, it rolls the
• 11:47
  dice. There is a 1 in 6 chance that it will activate
• 11:51
  a destructive branch of code and execute the
• 11:53
  command RM -RF. And for anyone unfamiliar with
• 11:57
  Linux commands, RM -RF is the nuclear option.
• 12:01
  It's bad. It's very bad. It forcibly and recursively
• 12:05
  deletes everything on the hard drive. It wipes
• 12:08
  the operating system, the proprietary AI models,
• 12:11
  the historical data. I mean, everything is gone
• 12:13
  in seconds. This completely changes the narrative.
• 12:16
  In the first story with the web developers, we
• 12:17
  had a digital burglar looking for a quiet, lucrative
• 12:20
  payout. Right, stealing keys. Exactly. But this.
• 12:23
  This is a burglar who breaks into your server
• 12:25
  room, ignores the expense of hardware, checks
• 12:28
  the passport sitting on the desk, and based strictly
• 12:30
  on your nationality, decides whether to burn
• 12:32
  the building to the ground. Yeah, that's exactly
• 12:34
  what it is. Why are open source AI tools suddenly
• 12:37
  becoming the delivery mechanism for geopolitical
• 12:39
  arson? Because AI development tools sit at the
• 12:42
  absolute nexus of highly valuable intellectual
• 12:45
  property and massive computing scale, compromising
• 12:48
  a single machine learning engineer's workstation
• 12:51
  can grant an attacker access to a company's entire
• 12:54
  cloud cluster. So it's the perfect entry point.
• 12:57
  Exactly. And when you inject geofenced hard drive
• 13:00
  wipers into that environment, it signals a massive
• 13:03
  shift in motivation. These aren't financially
• 13:06
  motivated cyber criminals looking for a quick
• 13:08
  ransom anymore. Right. These are highly sophisticated,
• 13:11
  likely state sponsored or state aligned actors
• 13:14
  who are willing to cause severe operational disruption
• 13:17
  and sabotage within very specific geopolitical
• 13:20
  hotspots. It is a stark reminder that the digital
• 13:23
  supply chain is an active battlefield right now.
• 13:25
  But all of this highly sophisticated geofenced
• 13:28
  malware is entirely useless without the physical
• 13:31
  processors to actually run the AI models. Right.
• 13:33
  The hardware. Yeah. And the control of that physical
• 13:36
  hardware brings us directly to the tension currently
• 13:38
  surrounding NVIDIA. Because the software war
• 13:41
  relies entirely on silicon. And the flow of that
• 13:44
  silicon is currently dictated by world leaders
• 13:46
  and chief executives. Right. We are currently
• 13:49
  tracking this incredibly high stakes diplomatic
• 13:52
  and commercial trip to China being led by President
• 13:54
  Trump. He is bringing a heavyweight delegation
• 13:57
  of U .S. executives to meet with Chinese President
• 13:59
  Xi Jinping. It's a huge roster. Oh, massive.
• 14:03
  We are talking about Apple's Tim Cook, Tesla's
• 14:06
  Elon Musk and Boeing's Kelly Ortberg. And, you
• 14:10
  know, Ortberg has a lot riding on this. Boeing
• 14:12
  is actively hunting to secure a massive 500 jet
• 14:15
  order. for their 737 MSX, which would just be
• 14:19
  a transformative win for the company's current
• 14:20
  trajectory. Oh, totally. And the administration
• 14:23
  is really attempting to thread a very thin needle
• 14:25
  here. It's a diplomatic mission designed to project
• 14:28
  strength while simultaneously acting as a commercial
• 14:30
  mission to navigate trade tensions and secure
• 14:33
  vital economic wins for American companies. Right.
• 14:35
  But the biggest drama unfolding right now centers
• 14:38
  squarely on Nvidia CEO Jensen Huang. Oh, man,
• 14:41
  the reporting on this is giving the entire industry
• 14:44
  whiplash today. We have two completely conflicting
• 14:48
  narratives playing out in real time. On one side,
• 14:51
  you've got sources like CNBC and Blockonomy reporting
• 14:53
  that Jensen Huang was specifically not invited
• 14:56
  on this trip. And this is despite Huang publicly
• 14:59
  stating it would be a quote, great honor to attend.
• 15:01
  And if he is excluded, the implications are massive.
• 15:05
  NVIDIA's most advanced AI processors have faced
• 15:08
  incredibly strict U .S. export restrictions to
• 15:10
  China for years. Right. They haven't been able
• 15:12
  to sell their best stuff. Exactly. In response,
• 15:15
  NVIDIA poured resources into engineering a specific,
• 15:18
  slightly throttled chip, the H200, designed specifically
• 15:21
  to comply with those exact restrictions. And
• 15:23
  stay within the rules. Yeah. They did that to
• 15:25
  maintain access to what Huang estimates is a
• 15:27
  $50 billion annual AI market in China. So if
• 15:31
  he has left off the manifest for this diplomatic
• 15:34
  flight it strongly suggests those efforts are
• 15:36
  totally stalled and Washington is just not ready
• 15:39
  to thaw the hardware freeze and We saw the market
• 15:42
  instantly react to that exact rumor Nvidia stock
• 15:45
  took a point seven percent dip in pre -market
• 15:48
  trading just on the whisper that he wasn't going
• 15:51
  Yeah, but then on the exact flip side you have
• 15:54
  a report from Inves stating that Huang was in
• 15:57
  fact invited alongside the CEO of Exxon. And
• 16:00
  if the Inves report proves true, it sends the
• 16:02
  exact opposite geopolitical signal. Right. It
• 16:06
  would indicate that Washington actually recognizes
• 16:08
  the economic necessity of preserving AI chip
• 16:11
  access to the global market, and they are actively
• 16:14
  managing the relationship to keep NVIDIA dominant
• 16:17
  on the world stage. So they might even be looking
• 16:19
  to negotiate the easing of certain export controls.
• 16:22
  Exactly. It is a high stakes game of telephone.
• 16:24
  But why should you, the listener, care whether
• 16:27
  one billionaire gets a seat on a plane to Beijing?
• 16:30
  Because this directly impacts global inflation,
• 16:32
  it impacts the health of tech -heavy retirement
• 16:34
  portfolios, and it dictates the underlying cause
• 16:37
  of almost every smart device we use. It ripples
• 16:39
  through everything. It really does. Nvidia has
• 16:42
  a massive earnings call coming up on May 20th.
• 16:45
  Analysts are expecting them to report an astonishing
• 16:47
  $78 .6 billion in revenue. If that $50 billion
• 16:51
  Chinese market is permanently locked down or
• 16:54
  suddenly opened up, those numbers and the broader
• 16:57
  tech market will just swing violently. That is
• 16:59
  the reality of the hardware layer. You have these
• 17:02
  globally interconnected software ecosystems where
• 17:05
  Developers are unwittingly downloading self -replicating
• 17:08
  worms and triggering geo -fenced hard drive wipers,
• 17:12
  right? And at the exact same time, governments
• 17:14
  are fiercely locking down the physical boundaries
• 17:16
  of the processors that make it all run. It is
• 17:19
  a massive collision course. The code has no borders,
• 17:22
  but the silicon is guarded by militaries and
• 17:24
  trade embargoes. We are quite literally holding
• 17:27
  our breath to see who steps off that plane in
• 17:29
  Beijing because it will set the tone for the
• 17:31
  next decade of technological development. Every
• 17:33
  single layer of the tech stack from the open
• 17:35
  source code library to the physical microchip
• 17:37
  is heavily contested territory right now. Before
• 17:40
  we wrap up, a quick thanks to Old Glory, an iconic
• 17:42
  music and sports fan merch store, perfect for
• 17:45
  the techie in your life. Use promo code SNARFL
• 17:48
  for 15 % off at oldglory .com. So to tie all
• 17:53
  of these threads together, We are witnessing
• 17:55
  an unprecedented level of sophistication in how
• 17:58
  our digital infrastructure is being targeted.
• 18:00
  The stakes have dramatically escalated from silent
• 18:03
  data exfiltration to targeted politically motivated
• 18:07
  destruction. And the physical backbone of that
• 18:09
  entire digital world remains locked in a deeply
• 18:12
  complex diplomatic standoff between global superpowers.
• 18:16
  Which leaves us with a pretty heavy thought to
• 18:18
  ponder as you go about your week. If open source
• 18:21
  coding libraries, the very foundation of how
• 18:23
  we build the modern internet, can now be weaponized
• 18:25
  to wipe servers based strictly on the country
• 18:27
  you happen to live in, will the future of software
• 18:29
  development require digital passports and government
• 18:31
  background checks just to write a simple line
• 18:34
  of code? That is a fascinating and deeply unsettling
• 18:36
  question. I mean, we might literally be watching
• 18:38
  the end of the open internet as we know it. We're
• 18:40
  here to help at Snarful, so reach out on our
• 18:42
  website if you have any more questions. That's
• 18:48
  a wrap on today's top stories. Remember, tech
• 18:50
  doesn't have to be complicated, and we're here
• 18:53
  to help connect the dots. If you've got questions
• 18:56
  or want to learn more, visit us at snarful .com.
• 18:59
  Thanks for listening, and we'll catch you next
• 19:01
  time.