Latest / Tech Talks With Kinsoft / Last Week in Tech – Apple's Gemini-Powered Siri, ShinyHunters Hit PeopleSoft, and the AI-Chip Land Grab

Transcript

Last Week in Tech – Apple's Gemini-Powered Siri, ShinyHunters Hit PeopleSoft, and the AI-Chip Land Grab · Tech Talks With Kinsoft

• 0:00
  Right now, your smartphone is getting smart enough
• 0:03
  to organize your life just by reading the context
• 0:06
  of what's on your screen. But at the exact same
• 0:09
  time... A forgotten 10 -year -old piece of human
• 0:12
  resources software just brought an entire university
• 0:15
  system to a grinding halt. Yeah, it's a staggering
• 0:19
  contradiction, honestly. We are watching this
• 0:21
  bleeding edge of futuristic technology just collide
• 0:24
  head -on with the rusted -out pipes of legacy
• 0:27
  IT infrastructure. Right. And, well, welcome
• 0:31
  to a brand -new Deep Dive. We have a completely
• 0:33
  fascinating stack of sources to go through today,
• 0:36
  covering the week of June 8th to the 14th. It's
• 0:39
  a busy week. It really is. And the mission for
• 0:41
  today is to explore that wild whiplash between
• 0:44
  those two realities. So for you listening, whether
• 0:46
  you're a business leader trying to figure out
• 0:47
  how to safely evaluate new AI tools for your
• 0:50
  team, or you're just an insanely curious learner
• 0:53
  wanting to know what is actually happening under
• 0:55
  the hood of the devices you use every day. Which
• 0:58
  is everyone right now. Exactly. This Deep Dog
• 1:00
  is going to arm you with the essential knowledge
• 1:02
  you need this week. And, you know, when you look
• 1:04
  closely at the facts across... all of our sources
• 1:07
  this week, there is a very clear overarching
• 1:10
  theme, one that connects the glittering AI updates
• 1:13
  to the massive cyber attacks. And that theme
• 1:16
  is supply chains. OK, let's untack this, because
• 1:20
  when most people hear supply chains, they think
• 1:22
  of cargo ships and warehouses, right? Yeah. So
• 1:25
  how does that apply to software? Well, we really
• 1:28
  need to think about digital supply chains. What
• 1:30
  we're seeing in the sources today is this deep
• 1:32
  connection between the data supply chains that
• 1:35
  power our cutting edge AI assistance and the
• 1:38
  well, the criminal supply chains operated by
• 1:41
  modern ransomware gangs. Oh, wow. Because whether
• 1:44
  you're a tech giant trying to serve up the smartest
• 1:46
  artificial intelligence or you're a cyber criminal
• 1:48
  trying to extort millions of dollars, your success
• 1:51
  depends entirely on knowing exactly who holds
• 1:54
  your data. Right. how it flows from point A to
• 1:56
  point B, and exactly where a backdoor was left
• 1:59
  open along the way. I love that framing. So let's
• 2:02
  just jump right into the front end of that digital
• 2:04
  supply chain, starting with the June 8th Apple
• 2:07
  Worldwide Developers Conference. Siri has finally
• 2:11
  been overhauled. We've been waiting for this
• 2:14
  integration, and it's now branded as Siri AI,
• 2:17
  running via Apple Intelligence. The demos they
• 2:21
  showed off were a massive leap forward from the
• 2:24
  basic voice commands we're used to. Yeah. The
• 2:26
  shift here is moving from a discrete tool that
• 2:29
  just like sets timers to an intelligence layer
• 2:32
  that is deeply embedded into the operating system
• 2:35
  itself. It understands the context. Exactly.
• 2:38
  It understands the context of your entire digital
• 2:40
  life. Yeah. I mean, you can tell Siri to find
• 2:43
  a specific photo of a person from years ago and
• 2:46
  drop it into a shared family album all by voice.
• 2:48
  You can ask it to. remind you the exact minute
• 2:51
  a ticket lottery opens up. You can even take
• 2:53
  actions based on what's currently displayed on
• 2:55
  your screen, which implies it's constantly analyzing
• 2:58
  your visual workflow. Plus, they rolled out a
• 3:00
  new standalone Siri app that actually remembers
• 3:03
  the context of your past conversations. And they're
• 3:06
  baking this intelligence directly into Spotlight
• 3:09
  Search on the Mac. Here is where the architecture
• 3:13
  of this system gets genuinely surprising. The
• 3:16
  underlying model infrastructure. Yes. Apple confirmed
• 3:19
  that this new highly capable Siri is actually
• 3:23
  powered in part by Google's Gemini model. Yeah.
• 3:26
  It's happening through Apple's Cloud Foundation
• 3:28
  model setup. And I just I have to push back on
• 3:31
  this strategy because it. It feels so completely
• 3:34
  jarring for this specific company. It did seem
• 3:37
  a bit out of character at first glance. Right.
• 3:39
  I mean, Apple built its entire modern brand identity
• 3:42
  on the concept of total privacy and vertical
• 3:44
  integration. They designed their own silicon.
• 3:46
  They write their own software. Leaning on their
• 3:48
  biggest rival's AI for their flagship assistant.
• 3:51
  Yeah. It's like finding out the secret ingredient
• 3:53
  in a Michelin star vegan restaurant's signature
• 3:55
  dish is actually sourced from a fast food burger
• 3:58
  joint. That is a phenomenal analogy. And the
• 4:01
  irony is on. unavoidable for sure. But if we
• 4:04
  look at the mechanics of how I actually functions.
• 4:07
  this decision makes perfect sense how so well
• 4:11
  apple is basically utilizing a peered supply
• 4:13
  chain for user queries so when you ask siri to
• 4:18
  do something highly personal like finding a photo
• 4:20
  or summarizing an email that processing happens
• 4:23
  locally on the device because that's where your
• 4:25
  sensitive data lives right exactly or it goes
• 4:28
  to apple's proprietary private cloud keeping
• 4:30
  the personal data locked down but when you ask
• 4:34
  siri a broad complex question about world history
• 4:36
  or, you know, ask it to generate a complicated
• 4:38
  piece of code. It is more juice. Yeah. Apple's
• 4:41
  internal models just might not have the sheer
• 4:43
  breadth of general knowledge required to answer
• 4:45
  accurately without hallucinating. Building a
• 4:48
  model that knows everything about everything
• 4:50
  requires scraping the entire Internet. Which
• 4:52
  Google has already done. Exactly. Google has
• 4:54
  already done that with Gemini. So Apple basically
• 4:56
  acts as a secure router. They securely hand off
• 4:59
  that specific non -personal query to Google's
• 5:02
  infrastructure. So Apple is essentially admitting
• 5:04
  that right now, Acting as a broker for the best
• 5:08
  AI is safer than trying to build the entire encyclopedia
• 5:11
  themselves. Pretty much, yeah. What does this
• 5:14
  mean for the listener who's trying to implement
• 5:16
  AI in their own workflow? It establishes a crucial
• 5:19
  baseline. If a company with the virtually unlimited
• 5:22
  resources of Apple cannot build a completely
• 5:25
  self -contained, perfect AI ecosystem in -house,
• 5:29
  businesses shouldn't expect to either. You're
• 5:31
  always going to be relying on third parties.
• 5:33
  Right. So the question you must ask when evaluating
• 5:35
  any AI tool is, where is this model actually
• 5:39
  running? You might purchase an enterprise license
• 5:41
  from company X, but your proprietary business
• 5:44
  data might be getting routed and processed on
• 5:47
  company Y. Oh, that's a huge blind spot. It is.
• 5:50
  You have to map out that data supply chain before
• 5:52
  you deploy the tool. And the reason Apple has
• 5:54
  to hand off those massive general knowledge queries
• 5:57
  to Google is because of the staggering computing
• 5:59
  power required to run them, which seamlessly
• 6:02
  connects to the next massive shift in our sources.
• 6:06
  The physical hardware supply chain. We are watching
• 6:09
  a multi -billion dollar land grab for AI chips.
• 6:12
  The physical iron underneath all of this digital
• 6:15
  magic, right? Because you can't build a software
• 6:18
  supply chain without the hardware to support
• 6:20
  it. Right. Alphabet Google's parent company.
• 6:23
  has officially started renting out its eighth
• 6:25
  -generation custom AI chips, known as TPUs, and
• 6:29
  they are operating at an unbelievable scale.
• 6:32
  Massive. They formed a $5 billion joint venture
• 6:35
  with Blackstone specifically to lease server
• 6:39
  capacity to outside companies like Anthropic
• 6:42
  and Meta. We should probably clarify what a TPU
• 6:44
  actually is just because it explains why this
• 6:46
  hardware race is so fierce right now. Yeah, go
• 6:48
  for it. So a TPU is a tensor processing unit.
• 6:51
  For years, the tech industry relied on GPUs graphics
• 6:53
  processing units to train AI. Because GPUs were
• 6:56
  really good at doing many simple math problems
• 6:58
  at the same time. Right, which was originally
• 7:00
  needed for rendering video game graphics. But
• 7:03
  a TPU is custom designed from the ground up by
• 7:05
  Google specifically for the complex neural network
• 7:07
  math that drives machine learning. It strips
• 7:10
  away all the legacy graphics architecture. to
• 7:12
  focus purely on AI efficiency. Wow. And Google
• 7:15
  isn't the only giant building custom silicon
• 7:19
  here. Microsoft just rolled out its second generation
• 7:22
  Maya accelerator, the Maya 200, which is specifically
• 7:26
  built to run their co -pilot and open AI workloads.
• 7:29
  Now, the sources do point out that the vast majority
• 7:32
  of Azure AI still runs on NVIDIA hardware. NVIDIA
• 7:36
  spent this exact same week tightening its grip
• 7:39
  on the hardware supply chain across Asia. They're
• 7:41
  still the king right now. Right. But the trend
• 7:43
  is clear. Hyperscalers like Google and Microsoft
• 7:46
  are desperately trying to build their own chips
• 7:48
  to escape NVIDIA's absolute dominance of the
• 7:51
  market. Yeah, the margins on NVIDIA hardware
• 7:52
  are astronomical and the hyperscalers want to
• 7:54
  control their own destiny. But this is where
• 7:56
  I get confused by the corporate strategy here.
• 7:59
  If Microsoft and Google are spending billions
• 8:01
  of dollars in R &D to finally build their own
• 8:03
  custom chips and gain an edge. Yeah. Why is Google
• 8:06
  immediately turning around and renting those
• 8:08
  eighth generation TPUs out to direct competitors
• 8:10
  like Meta and Anthropic? I mean, if you finally
• 8:14
  have the best engine, why rent it to the other
• 8:16
  race cars? Why not hoard that advantage? It fundamentally
• 8:19
  comes down to the economics of hyperscale cloud
• 8:22
  computing. Okay. The infrastructure costs associated
• 8:25
  with building these data centers, you know, the
• 8:27
  physical land, the custom silicon, the massive
• 8:30
  cooling systems, the energy grids, they're so
• 8:33
  astronomically high that no single company can
• 8:36
  absorb them purely for internal use. Not even
• 8:38
  Google. Not even Google. Plus, these chips depreciate
• 8:41
  quickly as... newer generations are invented.
• 8:44
  You cannot afford to let them sit idle for even
• 8:46
  a second. So by renting them out, Google turns
• 8:49
  a massive capital expense into a highly lucrative
• 8:53
  revenue stream. So they realize the real money
• 8:55
  isn't just in building the best AI. It's in being
• 8:58
  the landlord for everyone else's AI. Exactly.
• 9:00
  They control the foundation of the physical supply
• 9:03
  chain. And for you, the end user, this dynamic
• 9:07
  is incredibly positive. How so? This fierce competition,
• 9:10
  Google, Microsoft, and Nvidia all battling to
• 9:13
  offer the most efficient compute power will ultimately
• 9:16
  drive down the cost of processing. A diversified
• 9:19
  hardware supply chain means better pricing and
• 9:21
  faster innovation for the software we actually
• 9:23
  use. And that rapidly expanding pool of rental
• 9:27
  hardware is exactly what is supercharging the
• 9:29
  companies trying to dethrone the current AI kings.
• 9:33
  Yeah. The cheaper and more available compute
• 9:35
  power becomes, the faster a challenger can iterate,
• 9:39
  which explains the explosive growth of a major
• 9:42
  contender highlighted in our sources, Anthropic's
• 9:45
  Claude. Anthropic has been uniquely positioned
• 9:48
  to take advantage of this hardware availability
• 9:50
  to scale their operations. The numbers in the
• 9:52
  report are just staggering. According to the
• 9:55
  data, Claude's worldwide web visit share was
• 9:57
  sitting around 8%. But then... They experienced
• 10:00
  a 300 % jump in a single quarter. Huge. Yeah,
• 10:04
  they surged from roughly 200 million visits in
• 10:06
  January to over 800 million in April. In just
• 10:10
  three months, they quadrupled their traffic.
• 10:12
  That kind of scaling isn't accidental. It points
• 10:14
  to a deliberate architectural and business shift.
• 10:17
  Anthropic realized that to capture the real value
• 10:20
  in this market, they couldn't just build a clever
• 10:22
  consumer chatbot. Right. They had to build highly
• 10:24
  efficient models with massive context windows
• 10:26
  that businesses could actually use to process
• 10:28
  hundreds of pages of internal documents at once.
• 10:31
  And they are aggressively capitalizing on that
• 10:34
  enterprise focus. The sources note that Anthropic
• 10:37
  just formalized a $100 million partner program.
• 10:40
  They're calling it the Partner Hub. And it includes
• 10:42
  a dedicated services track specifically aimed
• 10:45
  at helping businesses put Claude into production.
• 10:49
  This really solidifies for me that the AI -assisted
• 10:52
  market is no longer just about chat GPT. No,
• 10:55
  not at all. It feels like we have transitioned
• 10:56
  overnight from the concept car phase of AI where
• 11:00
  everyone is just marveling at the technology
• 11:02
  straight into mass -producing sedans for enterprise
• 11:05
  fleets. The novelty phase is officially over.
• 11:08
  And the key phrase in the sources that we need
• 11:10
  to analyze is putting it into production. OK,
• 11:13
  unpack that. Vendors like Anthropic are no longer
• 11:16
  just handing a business a web login for their
• 11:18
  employees to use. They are actively pushing to
• 11:21
  integrate their AI via APIs directly into a company's
• 11:25
  internal workflows. So they're essentially just
• 11:27
  embedding themselves right into. Into the daily
• 11:29
  operations. Yeah, exactly. Their customer service
• 11:31
  staff boards and their proprietary databases.
• 11:35
  And this serves as a critical warning for anyone
• 11:37
  listening. If your team is trialing AI right
• 11:40
  now, the marketplace of available tools is widening
• 11:43
  rapidly, and the pressure to integrate them will
• 11:45
  be intense. Oh, for sure. But when you put AI
• 11:48
  into production, you are actively stitching a
• 11:51
  new powerful entity into your internal data supply
• 11:54
  chain. Conversations about data handling, retention
• 11:57
  policies, and security access cannot be an afterthought.
• 12:01
  They must happen on day one before a single API
• 12:04
  key is generated. Security from day one. is such
• 12:07
  a crucial mindset. Because while the entire tech
• 12:10
  world is utterly mesmerized by integrating these
• 12:13
  cutting -edge screen -reading AI tools, the back
• 12:16
  door of the enterprise is being left wide open.
• 12:18
  Absolutely. We are building these highly complex,
• 12:20
  futuristic supply chains, but ignoring the rusted
• 12:23
  locks on the doors we already have. Which brings
• 12:25
  us to the dark, unglamorous reality of this week's
• 12:28
  news. The cyber -criminal element thrives on
• 12:30
  that exact distraction. There's an extortion
• 12:32
  crew known as Shiny Hunters. Google's threat
• 12:35
  intelligence team Mandiant actually tracks their
• 12:38
  activity under the designation UNC 6240. And
• 12:42
  from late May right up to June 9th, this crew
• 12:45
  exploited an unpatched flaw in Oracle PeopleSoft.
• 12:48
  For context, Oracle PeopleSoft is a classic foundational
• 12:51
  enterprise application used for human resources,
• 12:55
  finance, and student administration. Very old
• 12:57
  school. It is the definition of legacy infrastructure.
• 13:00
  And shiny hunters used a known unpatched flaw
• 13:03
  in this software to break into enterprise systems,
• 13:06
  steal vast amounts of data and demand extortion
• 13:09
  payments. The sources specifically note that
• 13:11
  universities were hit the hardest by this campaign.
• 13:14
  Right. On top of that, on June 10th, CISA, the
• 13:17
  U .S. Cybersecurity and Infrastructure Security
• 13:19
  Agency, added three more actively attacked vulnerabilities
• 13:23
  to its known exploited list. And, you know, when
• 13:25
  a vulnerability lands on the CISA known exploited
• 13:27
  list, it isn't theoretical. It means criminals.
• 13:29
  are actively weaponizing it in the wild right
• 13:31
  now to breach networks. But the contrast here
• 13:34
  is making my head spin. We literally just spent
• 13:37
  the first half of this deep dive talking about
• 13:39
  billions of dollars spent on custom silicon and
• 13:43
  AI models that can visually analyze a user's
• 13:46
  desktop. Yeah. And yet massive multimillion dollar
• 13:49
  institutions like universities are being completely
• 13:53
  taken down simply because someone didn't click
• 13:55
  update on an enterprise HR app. How does a sophisticated
• 13:59
  institution fall to something so basic? It is
• 14:02
  baffling, but it highlights the incredibly unglamorous
• 14:05
  nature of cyber defense. To understand why universities
• 14:08
  are hit so hard, you have to look at how they
• 14:09
  operate. Very decentralized. Incredibly decentralized
• 14:12
  IT departments, massive legacy user bases of
• 14:16
  students and alumni, and hundreds of web -facing
• 14:18
  portals. It makes tracking down every single
• 14:21
  instance of an old application a nightmare. Criminals
• 14:24
  do not need hyper -advanced AI to hack you. They
• 14:26
  don't need custom Google. TPUs. They just use
• 14:29
  the exact same boring old playbook they have
• 14:31
  relied on for over a decade. Walk us through
• 14:33
  that playbook. How is it actually happening?
• 14:35
  It's a highly mechanized four -step process.
• 14:38
  Step one, use automated scanners to scour the
• 14:41
  internet for an exposed internet -facing enterprise
• 14:44
  app. Step two, the moment a vulnerability is
• 14:49
  publicly announced, race to exploit it before
• 14:51
  the target's IT team has time to test and apply
• 14:54
  the patch. Just a race against the clock. Exactly.
• 14:57
  Step three, move laterally through the network
• 15:00
  to exfiltrate the most sensitive data possible.
• 15:03
  And step four, lock the systems and extort the
• 15:06
  victim. It is entirely opportunistic. They are
• 15:09
  simply looking for the weakest link in your software
• 15:11
  supply chain. Which means the technological defense
• 15:14
  against this massive threat is literally just
• 15:16
  patching. Yes. But if the defense is that conceptually
• 15:19
  simple. And these highly organized gangs are
• 15:22
  this relentless. How do we actually stop them?
• 15:24
  Law enforcement agencies cannot possibly run
• 15:26
  around playing whack -a -mole, forcing thousands
• 15:28
  of individual universities to update their HR
• 15:31
  software. No, they cannot patch the Internet
• 15:33
  by force, which is why global law enforcement
• 15:36
  has fundamentally shifted their strategy. They
• 15:39
  are no longer just going after the software vulnerabilities.
• 15:41
  They're attacking the business model itself.
• 15:44
  They're targeting the financial supply chain.
• 15:46
  Following the money. Our sources detail a massive
• 15:49
  operation on June 12th by Europol. They executed
• 15:52
  a major disruption of a cryptocurrency laundering
• 15:55
  service that was used specifically by ransomware
• 15:58
  gangs to wash their illicit profits. Huge win.
• 16:01
  And the scale of this operation is hard to comprehend.
• 16:06
  This single service had claimed an estimated
• 16:08
  €336 million. That is an industrialized scale
• 16:12
  of money laundering. €336 million of extorted
• 16:15
  money washed clean and integrated back into the
• 16:18
  legitimate economy. How does the service even
• 16:21
  begin to hide that much digital currency? Well,
• 16:23
  it requires a sophisticated mechanism to break
• 16:26
  the blockchain's chain of custody. Cryptocurrencies
• 16:28
  like Bitcoin feature public ledgers, meaning
• 16:31
  anyone can track a ransom payment from the victim's
• 16:33
  wallet to the hacker's wallet. Right. It's all
• 16:36
  public. Exactly. So to actually spend that money,
• 16:39
  criminals use laundering services that employ
• 16:42
  mixers or tumblers. How do those work? Basically,
• 16:45
  they take the illicit cryptocurrency from...
• 16:48
  hundreds of different crimes and they pool it
• 16:50
  together. Like dumping it all into one massive
• 16:52
  pot. Right. And then they break it into thousands
• 16:56
  of microtransactions, rapidly swap it across
• 16:59
  different types of cryptocurrency chains, and
• 17:02
  then deposit the equivalent value back to the
• 17:05
  criminals. Oh, wow. Yeah. It obfuscates the origin
• 17:07
  so the funds can be cashed out for fiat currency
• 17:10
  safely. And shutting down that Tumblr service
• 17:12
  is a massive blow. But of course, as one threat
• 17:15
  vector is disrupted, another immediately rises
• 17:18
  to take its place. Always. The sources also mention
• 17:20
  a relatively new extortion crew that has quickly
• 17:23
  climbed the ranks to become one of the most active
• 17:25
  ransomware gangs globally by victim count. And
• 17:29
  they call themselves the gentlemen. It is a striking
• 17:32
  departure from typical threat actor branding.
• 17:34
  It really is. I was noting the bizarre contrast
• 17:36
  in hacker naming conventions. On one hand, you
• 17:39
  have a flashy, almost comic book name like shiny
• 17:42
  hunters. Right. And on the other, you have the
• 17:44
  weirdly polite, sophisticated sounding the gentlemen.
• 17:49
  But. Regardless of what they call themselves
• 17:51
  or what branding they adopt, they're all operating
• 17:54
  to feed a multi -hundred million euro... criminal
• 17:57
  ecosystem. Yeah. The naming conventions are just
• 17:59
  marketing. What is fascinating here is how clearly
• 18:03
  this Europol bust illustrates the core economics
• 18:06
  of ransomware. Yeah. Cutting that cryptocurrency
• 18:08
  money pipeline is the absolute most effective
• 18:11
  way to take the profit incentive out of cybercrime.
• 18:14
  If the gangs cannot wash the money, they cannot
• 18:16
  spend the money and the entire operation collapses.
• 18:19
  Which is the ultimate goal. Exactly. For the
• 18:21
  listener, the enduring lesson here is that the
• 18:23
  cast of characters will constantly rotate from
• 18:25
  shiny hunters to the to whoever emerges next
• 18:27
  month. But the exploitation of unpatched systems
• 18:31
  and the underlying financial motivations remain
• 18:33
  exactly the same. It truly all comes back to
• 18:36
  the ply chains. So to seamlessly wrap up our
• 18:39
  journey today, we have explored how Apple's brand
• 18:42
  new context -aware Siri AI evolution is actually
• 18:46
  relying on... Google's foundation model supply
• 18:48
  chain to handle complex queries. We dug into
• 18:51
  the multi -billion dollar hardware race where
• 18:54
  giants like Google and Microsoft are building
• 18:57
  custom silicon -like TPUs and immediately renting
• 19:00
  them out to control the physical infrastructure
• 19:02
  of AI. We saw Anthropix Quad leverage that compute
• 19:06
  power to achieve a massive 300 % surge in traffic
• 19:09
  as they push to integrate AI directly into enterprise
• 19:12
  production. And we contrasted all of that glittering
• 19:16
  future tech with the gritty, enduring reality
• 19:18
  of unpatched Oracle software bringing down universities
• 19:22
  and multi -million euro pole takedowns designed
• 19:26
  to sever the financial supply chains of cybercrime.
• 19:29
  It is a complex web of information, but seeing
• 19:31
  how the physical hardware, the AI software, and
• 19:33
  the criminal networks all intersect gives us
• 19:36
  a much clearer picture of the landscape. Absolutely.
• 19:38
  So for you listening, your immediate actionable
• 19:40
  takeaways for this week are crystal clear. First,
• 19:43
  know exactly what legacy enterprise apps you
• 19:45
  have exposed to the open internet right now.
• 19:47
  Have someone check the cease and unexploited
• 19:49
  list today and prioritize that patch queue. Second,
• 19:52
  when you are evaluating all these incredible
• 19:54
  new AI tools for your team, ask the hard questions
• 19:58
  about where your data actually lives and who
• 20:01
  else in the supply chain is processing it. And
• 20:03
  I want to leave you with a final provocative
• 20:05
  thought to mull over as you step away from this
• 20:07
  deep dive. Let's hear it. Throughout our discussion
• 20:09
  today, we have seen tech giants pooling immense
• 20:12
  resources to centralize unimaginable amounts
• 20:14
  of corporate data to feed their AI models and
• 20:17
  partner ecosystems. Meanwhile, highly organized
• 20:21
  ransomware syndicates like The Gentleman are
• 20:23
  perfecting the art of extracting data from forgotten,
• 20:26
  unpatched systems. Right. If the future of our
• 20:29
  business productivity requires us to feed our
• 20:32
  most intimate internal data into a complex, constantly
• 20:35
  shifting supply chain of rented AI chips and
• 20:38
  third -party partner programs, are we inadvertently
• 20:40
  building the ultimate, perfectly centralized
• 20:42
  payload for tomorrow's extortionists?