Latest / Tech Talks With Kinsoft / Last Week in Tech – OpenAI Retires GPT-4o, China-Linked Telco Espionage, and an Ivanti Zero-Day Hits Brussels

Transcript

Last Week in Tech – OpenAI Retires GPT-4o, China-Linked Telco Espionage, and an Ivanti Zero-Day Hits Brussels · Tech Talks With Kinsoft

• 0:00
  Imagine waking up tomorrow, grabbing your coffee,
• 0:02
  opening your laptop, and discovering that the
• 0:05
  digital foundation your entire company runs on
• 0:07
  has just, well, vanished. Right. Just completely
• 0:12
  gone. Exactly. It hasn't been hacked. It isn't
• 0:15
  temporarily down for maintenance. It is just
• 0:18
  gone, erased. And today we're looking at a week
• 0:21
  where that exact scenario actually played out
• 0:24
  for thousands of businesses globally. We are
• 0:27
  unpacking why the cybersecurity perimeters we
• 0:30
  blindly trust to protect our workflows are cracking
• 0:34
  right down the middle. Yeah, and it forces a
• 0:36
  total recalibration of how we approach enterprise
• 0:38
  architecture. I mean, we operate under this illusion
• 0:40
  of permanence. We deploy a tool, we build our
• 0:43
  defenses, and we just assume that because we
• 0:45
  paid for them or, you know, because they have
• 0:47
  a big shiny vendor logo on them, that the ground
• 0:50
  beneath us is solid. Right, but it's not. Not
• 0:52
  at all. The telemetry we are seeing right now
• 0:54
  proves that the ground is completely fluid. Which
• 0:56
  brings us to today's deep dive. We are tearing
• 0:59
  into a really dense, incredibly revealing update
• 1:02
  from Kinsoft Tech Talks. And this is dated February
• 1:06
  16th, 2026. The mission today is to examine a
• 1:11
  chilling theme hidden in this week's tech news,
• 1:13
  which is the severe systemic danger of the misplaced
• 1:17
  trust we put into our digital infrastructure.
• 1:19
  Oh, absolutely. And nowhere is that infrastructure
• 1:21
  evaporating faster than in the realm of artificial
• 1:24
  intelligence. So the Kinsoft update kicks off
• 1:28
  with OpenAI initiating this massive, unceremonious
• 1:31
  house cleaning. They retired GPT -4 .0, GPT -4
• 1:34
  .1, the 04 Mini, and the original GPT -5 models
• 1:37
  from the chat GPT API, essentially forced everyone
• 1:40
  onto GPT -5 .2 overnight. Yeah. And on the surface,
• 1:43
  if you are just, I don't know, a casual user
• 1:45
  asking a chatbot to write a grocery list, you
• 1:47
  probably didn't even notice. Right. You just
• 1:49
  open the app and it works. Exactly. But in the
• 1:51
  enterprise environment, deprecating models like
• 1:53
  that is an architectural earthquake. It's a huge
• 1:55
  deal. Okay. Let's unpack this because if you
• 1:57
  were building a modern business, you don't just
• 1:59
  plug an AI model into your workflow. You weave
• 2:02
  it into the very fabric of your operations. You
• 2:04
  really do. You spend months engineering prompts,
• 2:07
  tuning system instructions. You know, formatting
• 2:10
  JSON parsers to handle the exact latency, the
• 2:14
  token limits, and the specific hallucination
• 2:16
  quirks of, say, GPT -4 .0. Building your critical
• 2:20
  business infrastructure directly on top of a
• 2:23
  specific hard -coded AI model like that is essentially
• 2:26
  like building a massive corporate skyscraper
• 2:28
  on a rented foundation. That is a great way to
• 2:31
  put it. Right. Because the landlord can legally
• 2:33
  swap the concrete out for gravel while you sleep.
• 2:36
  You wake up. And your customer service platform
• 2:38
  is suddenly throwing 500 level air codes because
• 2:41
  the API endpoint you rely on simply no longer
• 2:44
  exists. Yeah. And the Kinsoft report highlights
• 2:46
  a fascinating and, well, somewhat terrifying
• 2:49
  data point regarding exactly that. They noted
• 2:51
  that leading up to the deprecation, barely a
• 2:53
  tenth of a percent of users were still hitting
• 2:55
  the GPT -40 endpoints daily. A tenth of a percent.
• 2:58
  I mean, that sounds small. It does sound small,
• 3:00
  right? But in the context of OpenAI's massive
• 3:03
  global user base, a tenth of a percent is still.
• 3:07
  thousands of enterprise applications, automated
• 3:09
  scripts, and customer -facing interfaces all
• 3:12
  breaking at the exact same time. Exactly. Thousands
• 3:15
  of IT teams having the worst morning of their
• 3:17
  lives. And the failure here isn't on OpenAI,
• 3:20
  right? It's an enterprise architecture failure.
• 3:23
  When you hardwire a complex system to a single
• 3:26
  model version, you are over -optimizing for the
• 3:29
  present. and just totally ignoring the reality
• 3:31
  of cloud infrastructure. You're betting your
• 3:34
  uptime on an external vendor's deprecation schedule.
• 3:37
  But what is the alternative? I mean, if an engineering
• 3:39
  team spends six months getting the tone and formatting
• 3:42
  of GPT -40 exactly right for an automated legal
• 3:45
  review tool, they can't just flip a switch to
• 3:48
  GPT -5 .2. No, of course not. The output drift
• 3:51
  alone would break the downstream processing.
• 3:53
  The JSON structures might change. The verbosity
• 3:55
  changes. You can't just, you know, swap the API
• 3:58
  key and go to lunch. Right, which is why you
• 4:00
  don't hardwire it in the first place. You implement
• 4:02
  an AI abstraction layer. Okay, an abstraction
• 4:04
  layer. Tell me more about that. So, you need
• 4:06
  middleware that acts as a translator between
• 4:09
  your core application and the LOM vendor. Instead
• 4:12
  of your app asking GPT -4o directly to analyze
• 4:15
  a document, your app asks your internal gateway.
• 4:19
  That gateway handles the routing, the prompt
• 4:21
  translation, and the error handling. Oh, I see.
• 4:24
  Yeah, so if a model gets deprecated, you just
• 4:26
  update the abstraction layer to map to the new
• 4:28
  model, and your core application never knows
• 4:30
  the difference. The lesson here is that you simply
• 4:33
  cannot trust your tools to remain static. The
• 4:36
  infrastructure will be deprecated right out from
• 4:38
  under you. The deprecation of those AI hodls
• 4:41
  absolutely proves that trusting external vendors
• 4:44
  for stability is a massive operational risk.
• 4:46
  But that transitions us to an even more critical
• 4:49
  question. Which is? What happens when we apply
• 4:51
  that exact same blind trust to our own internal
• 4:54
  perimeters? I mean, if an AI tool you choose
• 4:56
  to trust can vanish, what happens when a security
• 4:58
  wall you have to trust gets shattered from the
• 5:00
  outside? Yeah, and this is where the Kinsoft
• 5:02
  update shifts from a discussion about operational
• 5:05
  resilience into a genuine crisis of cybersecurity.
• 5:08
  It gets really dark, really fast. It does. The
• 5:12
  report details this relentless wave of attacks
• 5:14
  targeting edge devices, and specifically, they
• 5:18
  highlight critical zero -day vulnerabilities
• 5:20
  in Avanti's Endpoint Manager Mobile. And just
• 5:23
  to underscore the severity of this for anyone
• 5:25
  listening, Avanti Endpoint Manager Mobile isn't
• 5:28
  just some peripheral, nice -to -have tool. Not
• 5:31
  at all. For an enterprise, an MDM, a mobile device
• 5:35
  management server, is the absolute keys to the
• 5:38
  kingdom. It has administrative access to every
• 5:40
  company phone, tablet, and remote device in the
• 5:43
  fleet. It can wipe them. It can install apps.
• 5:46
  It tracks them. It's the nerve center. Exactly.
• 5:48
  And attackers use these Ivanti Zero Days to breach
• 5:51
  the European Commission and parts of the Dutch
• 5:53
  government. What does this all mean for the rest
• 5:56
  of us if massive government bodies are getting
• 5:58
  caught with their digital pants down? I mean,
• 6:01
  if highly funded organizations with sprawling
• 6:04
  security operation centers are getting breached
• 6:06
  straight through their edge perimeters, how is
• 6:08
  a mid -sized financial firm so to stand a chance?
• 6:11
  Well, what's fascinating here is the specific
• 6:13
  metric that the Kinsoft report focuses on regarding
• 6:16
  the European Commission breach. Brussels publicly
• 6:18
  stated that they detected and contained their
• 6:21
  incursion within about nine hours. Wait, OK,
• 6:23
  I have to push back. hard on that being framed
• 6:25
  as a positive outcome. Nine hours. I know. I
• 6:29
  know how it sounds. In an enterprise network,
• 6:31
  an attacker with hyper -privileged access from
• 6:34
  an MDM server can do an unbelievable amount of
• 6:38
  damage in nine hours. That isn't a silver lining.
• 6:42
  To me, that illustrates how terrifyingly fast
• 6:44
  an entire network can be compromised once a trusted
• 6:48
  edge device is broken. I totally hear that. And
• 6:50
  you're not wrong, but you have to look at it
• 6:52
  through the lens of modern threat architecture.
• 6:55
  Yes, nine hours is enough time to do damage,
• 6:57
  but you are comparing it to a hypothetical scenario
• 6:59
  of absolute prevention. Right. And that scenario
• 7:02
  just no longer exists. The days of the perfect,
• 7:05
  unbreachable firewall are over. So you're saying
• 7:07
  the perimeter is entirely porous now. Basically,
• 7:10
  yes. The perimeter is just a filter. It stops
• 7:12
  the automated noise, the low level stuff. But
• 7:14
  against a determined, well -resourced adversary
• 7:17
  wielding a zero day exploit, the wall will fail.
• 7:20
  It's just a matter of when. Exactly. And the
• 7:24
  historical dwell time for threat actors inside
• 7:27
  a network used to be measured in months, sometimes
• 7:30
  even a year. If an attacker has administrative
• 7:33
  access for a month, they map your active directory,
• 7:36
  they execute Kerberosting attacks to harvest
• 7:39
  service account credentials, they move laterally
• 7:42
  to your backup servers, and they plant these
• 7:45
  deep, deep persistence mechanisms. Yeah, they
• 7:48
  just dig in like a tick. Exactly. By the time
• 7:50
  you spot them, you don't just... have to kick
• 7:52
  them out. You have to rebuild your entire domain
• 7:54
  from scratch. Because they've poisoned the well
• 7:56
  completely. Precisely. So by containing the breach
• 7:59
  in nine hours, Brussels operated on an assumed
• 8:02
  breach mentality. They didn't just trust their
• 8:05
  Ivanti server to keep attackers out. They actively
• 8:08
  monitored the internal network traffic for anomalous
• 8:11
  behavior, assuming the firewall had already failed.
• 8:13
  Oh, I get it. So the nine hours is a win because
• 8:16
  it's not a month. Right. Nine hours means they
• 8:18
  stopped the lateral movement before the attackers
• 8:20
  could cement a permanent foothold. Fast detection
• 8:23
  is really the only viable defense mechanism we
• 8:25
  have left. That makes the next section of the
• 8:28
  Kinsoft report even more alarming to me. Because
• 8:31
  Brussels had a loud nine -hour firefight. They
• 8:34
  saw the anomaly, they responded, they contained
• 8:36
  it. But the update immediately contrasts that
• 8:39
  active defense with attackers who are playing
• 8:42
  a much quieter long -term game. Oh yeah, a very
• 8:44
  different kind of threat. We are moving from
• 8:46
  edge devices into the core of the telecommunications
• 8:50
  sector. The telco infrastructure breaches. This
• 8:53
  completely redefines the scale of the problem.
• 8:55
  It really does. The report details a campaign
• 8:58
  by a threat group known as UNC 388A6. And just
• 9:02
  to be completely clear with you listening, the
• 9:04
  source material explicitly notes that this is
• 9:06
  a China -linked espionage group. We are simply
• 9:09
  reporting the facts and the attribution exactly
• 9:11
  as they were documented in the Kinsloft update.
• 9:14
  We aren't taking any political stance here, just
• 9:16
  examining the mechanics of the attack. Right,
• 9:18
  keeping it focused on the tech. Exactly. So this
• 9:21
  group successfully compromised all four of Singapore's
• 9:24
  major telecommunications companies in a highly
• 9:26
  sophisticated, really long -running campaign.
• 9:29
  And the Kinsloft report places that... right
• 9:31
  alongside another major telco incident. The Dutch
• 9:34
  provider Odido disclosed a breach exposing the
• 9:37
  personal data of 6 .2 million customers. But
• 9:41
  these two incidents represent fundamentally different
• 9:45
  threat models, don't they? I mean, they don't
• 9:46
  even belong in the same category of attack. The
• 9:49
  Odido breach, exposing 6 .2 million records,
• 9:52
  is a massive data exfiltration event. It's horrible
• 9:56
  for privacy, but architecturally, it's a smash
• 9:58
  and grab. Yeah, it's noisy. Right. You back a
• 10:01
  truck up, blow the vault doors, scoop up the
• 10:03
  data, and trigger every single alarm in the building
• 10:05
  as you drive away. It is obvious, but compromising
• 10:08
  the core infrastructure of all four major telcos
• 10:11
  in Singapore. That isn't a bank heist. No, it's
• 10:14
  not. If we connect this to the bigger picture,
• 10:16
  you really have to understand why telcos are
• 10:18
  the ultimate target. They aren't just holding
• 10:20
  customer data. They are the physical pipes of
• 10:22
  the digital economy. Right. It's not breaking
• 10:24
  into the bank. It is compromising the city's
• 10:27
  entire water supply. You don't need to pick the
• 10:29
  locks on individual houses if you completely
• 10:31
  control the reservoir feeding the pipes. That's
• 10:33
  a perfect analogy. If you are sitting inside
• 10:35
  the telco infrastructure. You have access to
• 10:38
  the metadata, the routing tables, and potentially
• 10:40
  the SMS multi -factor authentication tokens of
• 10:43
  every single citizen and enterprise. utilizing
• 10:47
  that network. And this is where the how becomes
• 10:49
  so critical, especially compared to the Brussels
• 10:51
  incident. UNC 3886 didn't just phish an employee
• 10:56
  and log into an admin portal. They use specialized
• 10:58
  malware designed specifically for virtualized
• 11:01
  environments. They were deploying root kits that
• 11:03
  sit at the hypervisor level. We're below the
• 11:06
  operating system. Exactly. Below the OS. When
• 11:09
  malware is sitting on the ESXi hypervisor, the
• 11:12
  virtual machines running on top of it and the
• 11:14
  standard endpoint detection and response tools
• 11:16
  running inside those virtual machines literally
• 11:19
  cannot see it. Because they're looking inside
• 11:21
  the house, but the foundation is rigged. Yes.
• 11:23
  The antivirus thinks the system is clean because
• 11:25
  the underlying foundation of the server itself
• 11:27
  is compromised. This is why the assumed breach
• 11:30
  model is so incredibly difficult to execute perfectly.
• 11:33
  If your monitoring tools are running on a compromised
• 11:36
  hypervisor, they are just lying to you. That
• 11:38
  is terrifying. It is. This invisible threat allows
• 11:42
  state -sponsored actors to map critical infrastructure
• 11:45
  and prepare for future sabotage without ever
• 11:48
  tripping a single alarm. So we have hypervisor
• 11:51
  -level rootkits compromising national telcos
• 11:53
  and zero -days ripping through government -edged
• 11:56
  devices. For a lot of IT professionals listening
• 11:58
  to this, it's easy to feel a sense of detachment.
• 12:01
  You might think, you know, I don't run a national
• 12:05
  telecommunications grid and I'm not the European
• 12:06
  Commission. These ultra -sophisticated attacks
• 12:09
  don't apply to my daily operations. But Kinsoft
• 12:12
  does not let us off the hook there. Not even
• 12:15
  slightly. They bring the threat directly to the
• 12:17
  absolute most mundane everyday tool in the corporate
• 12:20
  arsenal. Yep, email. They warn that severe threats
• 12:23
  are walking straight through the front door of
• 12:25
  our trusted environment. The report highlights
• 12:28
  a malicious outlook at him. We were talking about
• 12:30
  email plugins. And this wasn't some sketchy executable
• 12:33
  file downloaded from a dark web forum. This malicious
• 12:36
  add -in was hosted and distributed directly through
• 12:39
  the official Microsoft store. And before it was
• 12:42
  finally detected and pulled, it was used to successfully
• 12:44
  compromise roughly 4 ,000 Microsoft Enterprise
• 12:47
  accounts. Which is just a catastrophic number.
• 12:50
  I mean, an Enterprise Microsoft account isn't
• 12:52
  just access to someone's inbox. It is the identity
• 12:55
  plane for the user. It's everything. It grants
• 12:57
  access to their SharePoint architecture, their
• 12:59
  Teams communications, internal financial documents,
• 13:03
  and potentially the whole corporate directory.
• 13:06
  Here's where it gets really interesting. How
• 13:08
  on earth does a piece of malware just bypass
• 13:10
  Microsoft's security and sit proudly in the official
• 13:14
  store? We are all conditioned from day one in
• 13:17
  corporate IT to trust the ecosystem. Right, the
• 13:20
  blue checkmark mentality. Yes. If Microsoft puts
• 13:23
  a checkmark next to an add -in on their official
• 13:25
  marketplace, we just assume it has been rigorously
• 13:28
  reverse -engineered, sandboxed, and purified
• 13:31
  by a team of elite cybersecurity engineers. But
• 13:34
  the Kinsoft report is incredibly explicit here.
• 13:37
  Proof that an application came from an official
• 13:39
  store is absolutely not the same thing as proof
• 13:41
  that it is safe. And this raises an important
• 13:43
  question about the psychology of enterprise security.
• 13:46
  How so? Well, we have this desperate desire to
• 13:49
  outsource our critical thinking. IT administrators
• 13:52
  are overworked. They're exhausted. So they rely
• 13:54
  on the platform owner, Microsoft, Google, Apple
• 13:57
  to do the vetting for them. But Fred actors understand
• 14:01
  exactly how the marketplace automated scanning
• 14:03
  works. They know the blind spots. Exactly. Microsoft
• 14:07
  scanners are looking for traditional malware
• 14:09
  signatures. They are looking for code that tries
• 14:12
  to inject malicious payloads into memory. or
• 14:16
  establish illicit command and control connections.
• 14:19
  But a malicious Outlook add -in doesn't need
• 14:21
  to do any of that, does it? No, not at all. It
• 14:23
  just uses standard, legitimate Microsoft APIs
• 14:26
  to ask the user for permission. It requests an
• 14:29
  OAuth token. Okay, let's drill into that, because
• 14:32
  the mechanism here is why this is so successful.
• 14:35
  When an employee clicks Accept, on one of these
• 14:38
  shiny, official -looking add -ins, they aren't
• 14:40
  giving the attacker their password. No, and honestly,
• 14:43
  the attacker doesn't even want their password.
• 14:45
  Why not? Because if they steal a password, they
• 14:47
  still have to bypass your multi -factor authentication.
• 14:49
  Ah, right, the MFA prompt. Exactly. Instead,
• 14:53
  the add -in asks for an OAuth token. You can
• 14:56
  think of an OAuth token like a digital hotel
• 14:59
  keycard. When you check into a hotel, you prove
• 15:02
  your identity at the front desk once. That's
• 15:04
  your password and your MFA. The desk verifies
• 15:07
  you, and they give you a keycard. From that point
• 15:09
  on, you just tap the card to open your door.
• 15:12
  You don't show your passport to the door every
• 15:14
  single time you want to get into your room. And
• 15:16
  these malicious add -ins are essentially tricking
• 15:19
  the user into handing a master keycard directly
• 15:22
  to the attacker. Precisely. The user clicks accept
• 15:25
  on a prompt that says something harmless, like
• 15:27
  allow this app to read your emails and access
• 15:29
  your files. Microsoft generates the oath token
• 15:32
  and hands it straight to the add -in. Game over.
• 15:34
  Yeah. The attacker now has persistent programmatic
• 15:37
  access to the user's entire Microsoft 365 environment.
• 15:41
  They don't need to hack the network perimeter.
• 15:43
  They don't need to defeat your MFA. They've been
• 15:46
  legally handed the keys by an employee who simply
• 15:49
  trusted the official Microsoft Store logo. Threat
• 15:52
  actors are totally weaponizing our misplaced
• 15:54
  trust in these ecosystems. Man, we have covered
• 15:57
  a massive amount of ground today, and the implications
• 16:00
  of this Kinsoft update are heavy. We've seen
• 16:02
  core AI models vanish without warning, forcing
• 16:06
  enterprises into these panicked architectural
• 16:08
  scrambles. We've watched critical zero -day exploits
• 16:11
  tear down the edge defenses of the European Commission.
• 16:14
  We've examined hypervisor -level espionage living
• 16:17
  silently inside foundational telco infrastructure.
• 16:21
  we've unpacked how malicious actors are legally
• 16:23
  bypassing MFA by poisoning trusted software marketplaces
• 16:27
  with oath token harvesters. It paints a very
• 16:30
  stark picture of the modern threat landscape
• 16:32
  for sure. But the Kinsoft Tech Talks update doesn't
• 16:35
  just leave the listener drowning in paranoia.
• 16:37
  There's a very concrete through line here. Internet
• 16:40
  facing edge devices and trusted ecosystem add
• 16:42
  -ins are the primary vectors where attackers
• 16:44
  are finding the most success right now. And it's
• 16:47
  precisely because they are the areas where organizations
• 16:49
  mistakenly assume the vendor has already solved
• 16:52
  the security problem. For you listening, if you
• 16:55
  are an IT director, a security architect, or
• 16:58
  a business owner trying to operationalize this
• 17:00
  intelligence, here is your concrete action plan.
• 17:03
  First, you have to assume breach. You cannot
• 17:05
  just rely on your firewalls or your MDM servers,
• 17:08
  as we saw with Brussels. Your survival depends
• 17:11
  entirely on how fast you can detect lateral movement
• 17:13
  inside your network. Are you actively hunting
• 17:16
  for anomalous Kerberos tickets or unusual Active
• 17:19
  Directory queries right now? If not, you need
• 17:23
  to be. Second, you must aggressively lock down
• 17:26
  user consent for enterprise applications. You
• 17:28
  simply cannot allow standard employees to grant
• 17:31
  OOOTH permissions to third -party add -ins, regardless
• 17:34
  of whether they come from the official Microsoft
• 17:36
  or Google store. You have to take that power
• 17:38
  away. You do. You have to implement administrative
• 17:40
  workflows, where every single requested integration
• 17:43
  is audited for permission scopes before the token
• 17:46
  is granted. And finally, regarding your workflows,
• 17:49
  stop hardwiring your operations to specific external
• 17:52
  models. models. Whether it is an LLM or a cloud
• 17:55
  API, build abstraction layers. Do not tie your
• 17:59
  skyscraper to a rented foundation that will inevitably
• 18:01
  shift. And as the Kinsoft report notes, if you
• 18:05
  are sitting there wondering exactly what your
• 18:06
  attack surface looks like right now, you cannot
• 18:09
  protect an environment you haven't fully mapped.
• 18:12
  They advise visiting kinsoft .com .au if you
• 18:15
  need comprehensive help assessing your exposure
• 18:17
  to these exact vectors. Yeah, and the host of
• 18:19
  the Kinsoft update signed off with a phrase that
• 18:21
  serves as the the perfect operating principle
• 18:23
  for this entire discussion. They said, stay patched,
• 18:26
  stay skeptical. It really is the only way forward.
• 18:29
  But looking at everything we've unpacked today,
• 18:31
  from the ephemeral nature of our AI tools to
• 18:32
  the deeply compromised state of hypervisors and
• 18:35
  official marketplaces, it forces a fundamental
• 18:37
  reevaluation of how we do business. We naturally
• 18:40
  want bedrock. We want to designate our firewalls
• 18:43
  and our software vendors as safe so we can stop
• 18:45
  worrying about them and just, you know, focus
• 18:48
  on our actual work. But the telemetry proves
• 18:50
  that safety is a total... Right. And I want to
• 18:52
  leave you with a final lingering thought to chew
• 18:54
  on regarding that illusion of safety. If the
• 18:58
  AI endpoints we build upon can just disappear,
• 19:01
  if the governmental perimeters we trust are breached
• 19:04
  in hours, and if the official stores we rely
• 19:07
  on are actively weaponized, well, is the very
• 19:10
  concept of trusting your IT infrastructure an
• 19:13
  outdated vulnerability? Moving forward, should
• 19:16
  enterprises completely abandon the idea of building
• 19:19
  a secure, unbreachable environment and instead
• 19:21
  design their operations assuming that the network
• 19:24
  is already compromised? If the spy is already
• 19:26
  living in the ceiling, how do you change the
• 19:29
  way you do business on the floor? That is the
• 19:31
  defining architectural question of the next decade.
• 19:33
  If you can't trust the foundation, you have to
• 19:36
  build resilience into every single transaction.
• 19:39
  Thank you so much for joining us for this deep
• 19:40
  dive into the Kinsoft intelligence. We hope it
• 19:43
  gave you the clarity to rethink your architecture,
• 19:45
  maybe a few aha moments about how these mechanisms
• 19:47
  actually work, and plenty to strategize over.
• 19:50
  Keep questioning your assumptions, keep exploring
• 19:52
  your networks, and we'll see you next time.