Latest / Tech Talks With Kinsoft / Last Week in Tech – OpenAI Pulls the Plug on Sora, a Court Backs Anthropic, and Hacktivists Claim a Lockheed Haul

Transcript

Last Week in Tech – OpenAI Pulls the Plug on Sora, a Court Backs Anthropic, and Hacktivists Claim a Lockheed Haul · Tech Talks With Kinsoft

• 0:00
  Imagine a major legacy automaker suddenly shutting
• 0:04
  down production on the most popular flashy sports
• 0:06
  car in the world. I mean, a car that everyone
• 0:10
  is just absolutely obsessed with. Right, the
• 0:12
  one that's on every magazine cover. Exactly.
• 0:14
  And they shut it down just so they can pull all
• 0:17
  their top engineers off the line to build a commercial
• 0:21
  tractor engine. You would think management had
• 0:23
  completely lost their minds. Yeah, you really
• 0:25
  would. But right now, the hottest artificial
• 0:28
  intelligence company on the planet is making
• 0:30
  exactly that kind of pivot. It's pretty wild
• 0:32
  to watch. It really is. So welcome to today's
• 0:36
  Deep Dive. We are thrilled you could join us
• 0:37
  because our mission today is to cut through the
• 0:40
  massive amount of noise in the tech world. And
• 0:42
  there is a lot of noise right now. Oh, tons.
• 0:44
  We're examining a really fascinating stack of
• 0:46
  excerpts from a recent tech briefing from Kinsoft.
• 0:49
  Specifically, their Tech Talks update dated March
• 0:53
  30th, 2026. Yeah, because we are just constantly
• 0:56
  bombarded with these loud, shiny headlines, like
• 0:59
  a massive cyber heist claim or some wildly entertaining
• 1:03
  new app, right? Right. And all our attention
• 1:05
  instantly goes there. Exactly. It's the ultimate
• 1:07
  distraction factor. So the overarching theme
• 1:10
  we really need to pull from this Kinsoft briefing
• 1:12
  is... Well, the illusion of the flashy versus
• 1:16
  the stark reality of infrastructure. Okay, let's
• 1:19
  unpack this, starting with that automaker analogy.
• 1:22
  Yeah, let's do it. Because one of the most surprising
• 1:25
  shifts in the Kinsoft briefing is about OpenAI
• 1:28
  shutting down its standalone Sora video app.
• 1:31
  Which is huge. It is. For those of you who might
• 1:34
  not follow the day -to -day product cycles, this
• 1:36
  was built up as this, like... TikTok -style app
• 1:38
  designed to generate these hyper -realistic,
• 1:41
  mind -bending videos for consumers. It was definitely
• 1:44
  the shiny sports car of the tech world. Right.
• 1:46
  And they're killing it. I mean, instead of focusing
• 1:49
  on entertaining consumer video feeds, they're
• 1:51
  taking all of that immense computing power and
• 1:53
  redirecting it. Toward world simulation and robotics
• 1:56
  research, which... Wait, why walk away from a
• 1:58
  consumer app that generates that much hype? It
• 2:01
  just seems counterintuitive. Because, well, the
• 2:03
  hype is kind of a dead end compared to the actual
• 2:05
  goal. Okay, how so? Well, Kinsoft's analysis
• 2:08
  points out that when a company with that much
• 2:11
  momentum actively chooses to kill a viral consumer
• 2:14
  product, it reveals their true endgame. Which
• 2:17
  is what exactly? The future of AI for these massive
• 2:20
  players isn't in entertaining us with endless
• 2:23
  streams of generated content. I mean, it isn't
• 2:25
  what Kinsoft calls a novelty feed. OK. The real
• 2:28
  prize is in developing agents. AI that can actually
• 2:31
  interact with and manipulate the physical world.
• 2:34
  So generating a photorealistic video of like
• 2:38
  a cat riding a skateboard on Mars is a fun parlor
• 2:41
  trick, but it's ultimately just pixels on a screen.
• 2:44
  Exactly. It doesn't fold your laundry. Right.
• 2:46
  It doesn't pack a box in a warehouse and it doesn't
• 2:48
  drive a delivery truck. And to do any of those
• 2:50
  physical tasks, an AI agent needs to really understand
• 2:54
  how the physical world actually works. Which
• 2:56
  is incredibly complicated. Super complicated.
• 2:59
  And this is where we... have to look at what
• 3:00
  it means to redirect compute. Compute being the
• 3:03
  raw processing power, right? Yeah, the incredibly
• 3:06
  expensive silicon brains of these operations.
• 3:10
  Up until now, these video generation models were
• 3:13
  basically learning how things looked. Okay, making
• 3:16
  things pretty. Right. But when you ask an AI
• 3:18
  to generate a video of, say, a glass falling
• 3:22
  off a table, to make it look realistic, the model
• 3:25
  essentially has to teach itself the concept of
• 3:27
  gravity. Oh, wow. I never really thought about
• 3:29
  it like that. Yeah, it has to learn that glass
• 3:31
  shatters, not bounces. It has to deeply understand
• 3:34
  spatial awareness and object permanence. So wait,
• 3:37
  they're taking the software engine that used
• 3:39
  to just draw pretty pictures and they're leveraging
• 3:42
  the underlying logic to build a massive physics
• 3:45
  engine. That's exactly it. They're using video
• 3:47
  generation as a backdoor to simulate physical
• 3:50
  reality. Which is exactly what you need to teach
• 3:52
  a robot how to navigate our world without crashing
• 3:55
  into one. Precisely. That is the pivotal transition
• 3:57
  here. They're moving from generating fun videos
• 4:00
  to modeling physical reality for robotics. That's
• 4:04
  a massive leap. It is a fundamental shift from
• 4:06
  being a consumer entertainment provider to building
• 4:11
  the foundational intelligence that will drive
• 4:13
  physical machines. So what does this all actually
• 4:17
  mean for the industry? Well, think about the
• 4:19
  scale of it. Does this imply open AI is essentially
• 4:23
  trying to become the underlying operating system
• 4:26
  for the physical world? I mean, that's certainly
• 4:29
  what it looks like. Because if you hold the keys
• 4:31
  to the foundational intelligence running commercial
• 4:34
  robotics, you aren't just a tech company anymore.
• 4:37
  your foundational infrastructure. You become
• 4:39
  the bedrock of the physical economy. Right. And
• 4:42
  the moment a tech company transitions from providing
• 4:44
  software to acting as foundational infrastructure,
• 4:47
  the rules of the game just change completely.
• 4:50
  The scrutiny skyrockets, I imagine. Oh, absolutely.
• 4:52
  Which actually leads perfectly into another major
• 4:55
  friction point the Kinsoft briefing highlights.
• 4:57
  But this one is regarding digital infrastructure.
• 5:00
  Yeah. Let's look at the facts laid out in the
• 5:02
  update about that. So a federal judge recently
• 5:05
  ruled that the Defense Department, the Pentagon,
• 5:08
  likely could not brand the AI company Anthropic
• 5:12
  a supply chain risk. Which is a huge development.
• 5:15
  It is. The judge went so far as to call the Pentagon's
• 5:19
  move probable unlawful retaliation for Anthropic's
• 5:22
  dissent. Yeah. And just to ground this for you
• 5:25
  listening, we need to be really clear here. Our
• 5:27
  goal today isn't to play judge and jury on who
• 5:30
  is right between the Pentagon and a major AI
• 5:32
  developer. Right. We're remaining strictly neutral
• 5:35
  on the politics of this. 100%. We're just digging
• 5:38
  into the Kinsov briefing to look purely at the
• 5:40
  mechanics of this dispute. Because the legal
• 5:42
  precedent it sets for how governments and tech
• 5:45
  companies interact is just massive. Exactly.
• 5:47
  So what does this all mean? When we think of
• 5:49
  a supply chain, we usually think of like... Cargo
• 5:52
  ships stacked with containers. Or a factory churning
• 5:55
  out microchips. Right. Or a fleet of delivery
• 5:57
  trucks. Yeah. How does an AI software company
• 6:00
  even become categorized as a supply chain risk
• 6:02
  to the United States military in the first place?
• 6:05
  Well, the mechanics are what make this so revealing.
• 6:08
  To really understand this dispute, we have to
• 6:10
  completely redefine what a supply chain actually
• 6:13
  is in 2026. Okay, redefine it how? A digital
• 6:17
  supply chain isn't about physical goods anymore.
• 6:19
  It's about the code. the algorithms and the data
• 6:23
  processing frameworks that modern organizations
• 6:25
  fundamentally depend on to function. So it's
• 6:28
  the invisible scaffolding. Exactly. Think about
• 6:30
  what a modern military requires today. They need
• 6:34
  advanced AI models to process massive amounts
• 6:37
  of raw intelligence, right? Right. And to optimize
• 6:40
  global logistics networks and manage real time
• 6:42
  cyber defenses. AI companies have basically become
• 6:45
  the sole suppliers of the raw cognitive material
• 6:48
  that runs modern institutions. Code is the new
• 6:50
  steel. Code is absolutely the new steel. But
• 6:53
  unlike a steel mill, which just melts iron and
• 6:56
  ships. out, no questions asked, an AI company
• 6:59
  has its own internal culture. Yeah, they have
• 7:01
  ethical guidelines, employee pushback. And as
• 7:04
  the briefing calls it, dissent. So you have this
• 7:06
  intense friction. The military needs absolute
• 7:09
  compliance and total reliability from its supply
• 7:12
  chain. But the supplier, the AI company, might
• 7:16
  decide that certain military applications violate
• 7:19
  their terms of service. Or their ethical boundaries,
• 7:22
  yeah. So when that tech company pushes back,
• 7:25
  How does a massive government entity respond?
• 7:27
  Well, in this specific case, the Pentagon attempted
• 7:30
  to label them a supply chain risk. Which is a
• 7:33
  heavy label to throw around. Very heavy. The
• 7:36
  implication is that if a supplier might cut you
• 7:38
  off because of an ethical disagreement, well,
• 7:40
  that supplier is a risk to your operational readiness.
• 7:44
  So you label them a risk to effectively cut them
• 7:46
  out of the infrastructure entirely and find someone
• 7:49
  more compliant. That's the idea. But then the
• 7:51
  federal judge steps in and it's the brakes. Calling
• 7:54
  it probable unlawful retaliation. Exactly. Which
• 7:58
  sets a fascinating boundary. It essentially says
• 8:00
  the government can't just weaponize a national
• 8:03
  security label to punish a tech company for having
• 8:05
  a dissenting voice. It really forces a highly
• 8:07
  complex relationship, doesn't it? Oh, incredibly
• 8:10
  complex. Our most critical defense systems are
• 8:14
  increasingly reliant on private tech companies.
• 8:17
  But those private companies are not silent, obedient
• 8:20
  factories. They are independent entities with
• 8:23
  their own agendas. Right. So navigating that
• 8:25
  digital supply chain is becoming just as complex
• 8:28
  legally and politically as it is technologically.
• 8:32
  It proves that the vulnerabilities in our digital
• 8:34
  infrastructure aren't just technical. They're
• 8:37
  deeply human and beautiful. Yeah, it's a massive
• 8:40
  legal standoff over theoretical risks in these
• 8:43
  high -level supply chains. But, you know, the
• 8:45
  irony here is incredibly thick. Oh, the irony
• 8:47
  is the best part. Because while the Pentagon
• 8:49
  and Anthropic are fighting it out in federal
• 8:51
  court over what constitutes a theoretical supply
• 8:54
  chain risk, the Kinsoft briefing points out that
• 8:57
  actual supply chains are actively bleeding data.
• 9:01
  Yeah, specifically the mundane third party vendors.
• 9:04
  Right. We are entirely distracted by the theoretical
• 9:06
  legal battles of the Titans while the house is
• 9:09
  quietly being robbed through the back door. It
• 9:11
  is the illusion of the flashy versus the reality
• 9:14
  of infrastructure all over again. Kinsoft actually
• 9:16
  uses the cybersecurity events of the week to
• 9:19
  provide like a masterclass in distinguishing
• 9:22
  between the noise we panic about and the reality
• 9:25
  we should actually be securing. OK, so let's
• 9:28
  start with the noise. The loud, flashy headline
• 9:30
  of the week. The big, scary one. Yeah. According
• 9:34
  to the briefing, a pro -Iran hacktivist group
• 9:37
  claimed they stole a staggering 375 terabytes
• 9:41
  of data from Lockheed Martin. Supposedly including
• 9:44
  blueprints for the F -35 fighter jet. Right.
• 9:48
  And they demanded a massive ransom. Again, for
• 9:50
  you listening, looking purely at the mechanics
• 9:52
  of cyberclaims rather than the geopolitical actors
• 9:55
  involved, this sounds like a terrifying Hollywood
• 9:58
  -level crisis. It sounds totally terrifying.
• 10:00
  But Kinsoft slaps a giant flashing warning sign
• 10:04
  on this. What kind of warning? They state really
• 10:06
  clearly that this is an unverified claim, and
• 10:08
  they emphasize that these specific types of hacktivist
• 10:10
  groups routinely exaggerate or just outright
• 10:13
  fabricate their triumphs. Here's where it gets
• 10:16
  really interesting. It's like a bank robber standing
• 10:18
  in the middle of the town square screaming that
• 10:21
  they just stole the gold reserves from the federal
• 10:23
  hall. Yeah, and everyone completely panics. The
• 10:25
  news vans show up, the stock market twitches.
• 10:28
  But in reality, someone just quietly picks the
• 10:31
  lock on the bank manager's personal gym locker.
• 10:34
  That analogy hits the underlying strategy perfectly.
• 10:37
  The claim is entirely designed to cause psychological
• 10:40
  chaos, not to reflect reality. Right. Let's actually
• 10:45
  look at the logistics of their claim for a second.
• 10:47
  Exfiltrating 375 terabytes of data is not like
• 10:51
  slipping a flash drive into your pocket. No,
• 10:53
  that is a truly astronomical amount of data.
• 10:56
  It is the equivalent of downloading nearly 100
• 10:58
  ,000 high -definition movies. Wow. Over a secure
• 11:02
  corporate network. Yeah. You cannot pull that
• 11:04
  much data out of a highly secure defense contractors
• 11:07
  network without lighting up every security dashboard
• 11:09
  like a Christmas tree. It would take massive
• 11:11
  bandwidth and significant time. Exactly. So the
• 11:14
  mechanics of the hack itself make the claim highly
• 11:16
  improbable. The attackers know that simply saying
• 11:19
  they have F -35 blueprints creates the illusion
• 11:22
  of total dominance. They weaponize the novelty
• 11:24
  feed against us. Which is terrifying in its own
• 11:27
  way. It is. Kinsoft uses this to hammer home
• 11:31
  a vital lesson, though. You must calmly assess
• 11:33
  the logistics of a leaked claim before reacting.
• 11:36
  Right. Because if an organization panics, shuts
• 11:39
  down systems, or engages with the attackers based
• 11:42
  purely on a boast. They are playing directly
• 11:45
  into the attacker's hands. The goal was disruption
• 11:47
  and the victim just provided it themselves. But
• 11:50
  we can't just ignore security altogether. Because
• 11:53
  while everyone is staring at the town square,
• 11:55
  listening to the guy screaming about 375 terabytes
• 11:58
  of stolen gold. The real thieves are busy elsewhere.
• 12:01
  Exactly. The Kinsop briefing lists three concrete,
• 12:05
  verified breaches that happened quietly elsewhere.
• 12:07
  And none of them look like a cinematic heist.
• 12:10
  No, they are remarkably, almost embarrassingly
• 12:13
  mundane. Let's run through them. First, the FBI
• 12:16
  director's personal email was reportedly hacked
• 12:18
  and leaked. Second, the massive anime streaming
• 12:21
  platform Crunchyroll was breached. But crucially,
• 12:24
  not through their main servers, but through a
• 12:26
  third -party support provider. And third, a French
• 12:29
  education ministry system exposing personal data
• 12:32
  of hundreds of thousands of staff members. The
• 12:34
  pattern there is just undeniable. We spend so
• 12:37
  much time terrified of someone hacking the mainframe
• 12:40
  of a defense contractor through a sheer brute
• 12:43
  force algorithm. Right. But the verified breaches
• 12:46
  show that high -profile targets are almost never
• 12:49
  compromised through a frontal assault on their
• 12:51
  strongest defenses. Attackers are incredibly
• 12:54
  pragmatic, aren't they? Highly pragmatic. Kinsoft
• 12:57
  identifies this pattern as targeting soft targets.
• 13:01
  Why spend months trying to break through a military
• 13:04
  -grade firewall when you can just find a side
• 13:07
  door that someone left unlocked? A side door
• 13:09
  like a personal email account. Right. You could
• 13:13
  have the most impenetrable classified government
• 13:15
  network in the world. But if the director of
• 13:18
  the FBI is checking a personal, less secure email
• 13:21
  account on their smartphone. That becomes the
• 13:23
  weak link. The attacker doesn't need to break
• 13:25
  into the FBI servers. They just need to trick
• 13:27
  someone into clicking a phishing link on their
• 13:29
  personal webmail. Exactly. And the exact same
• 13:32
  logic applies to the Crunchyroll breach, which
• 13:34
  is really the ultimate real -world example of
• 13:37
  the supply chain risk we were discussing earlier.
• 13:40
  Crunchyroll undoubtedly has incredibly robust
• 13:43
  security for their core streaming technology
• 13:45
  and their user database. Sure, they have to.
• 13:48
  But a massive company can't build every single
• 13:51
  tool from scratch. Third -party support providers
• 13:55
  to handle customer service tickets or manage
• 13:57
  internal HR portals. And those third -party providers
• 14:00
  become the soft target. Precisely. If an attacker
• 14:04
  knows Crunchyroll is a fortress, they don't attack
• 14:06
  the fortress. They go look at the IT ticketing
• 14:09
  software Crunchyroll buys from a smaller vendor.
• 14:11
  And if that smaller vendor has weak security,
• 14:13
  the hackers break into the vendor and just walk
• 14:16
  right through the established, trusted connection
• 14:19
  directly into Crunchyroll's data. Yeah. The third
• 14:22
  -party provider is the unglamorous, boring supply
• 14:24
  chain. It really doesn't make for a sexy headline,
• 14:27
  does it? No, but it is devastatingly effective.
• 14:29
  This is why the theoretical legal battle between
• 14:32
  the Pentagon and Anthropic is so critical. Right,
• 14:35
  because as organizations rely on increasingly
• 14:37
  complex webs of third -party software from IT...
• 14:41
  ticketing all the way to advanced AI cognitive
• 14:43
  engines, those connections become the primary
• 14:46
  vector for attack. It constantly comes back to
• 14:49
  the overarching theme from the Kinsoft briefing.
• 14:51
  We get terrified of the unverified 375 terabyte
• 14:55
  hack of fighter jet blueprints. We get completely
• 14:58
  distracted by the novelty feed. But the actual
• 15:01
  damage, the real loss of data, is happening because
• 15:04
  a third -party vendor forgot to patch their software.
• 15:08
  Or a high -ranking official reused a weak password
• 15:11
  on their personal email. It's the boring, everyday
• 15:14
  infrastructure that is crumbling while we stare
• 15:17
  at the shiny objects. You really have to separate
• 15:19
  the verified mechanics from the loud claims.
• 15:22
  Ignoring the noise and focusing relentlessly
• 15:24
  on those soft targets, the personal accounts
• 15:26
  and the third -party providers, is honestly the
• 15:28
  only way to actually secure a network. It requires
• 15:30
  a massive shift in perspective, but I think it
• 15:33
  brings us to a really clear synthesis of everything
• 15:35
  we've explored in this deep dive. Yeah, I agree.
• 15:38
  Today, the Kinsoft briefing showed us a technology
• 15:41
  landscape that is rapidly shedding the flashy
• 15:44
  in favor of the foundational. We saw OpenAI killing
• 15:47
  a viral consumer video app to redirect their
• 15:50
  massive computing power. Right, aiming it towards
• 15:53
  simulating the physical world, paving the way
• 15:55
  for advanced robotics. And we saw the intense...
• 15:59
  Precedent setting legal friction between AI companies
• 16:02
  and the government over what actually constitutes
• 16:05
  a digital supply chain risk. Exactly. And in
• 16:08
  the trenches of cybersecurity, we learned that
• 16:11
  the genuine threats aren't the headline grabbing
• 16:13
  unverified heists. They're the quiet, mundane
• 16:16
  vulnerabilities hiding in our personal devices
• 16:19
  and our third party vendors. The true shape of
• 16:21
  the tech world really is found in the infrastructure.
• 16:24
  It absolutely is. And so the practical takeaway
• 16:27
  for you, the listener, comes straight. Straight
• 16:28
  from the sign -off of the Kinsoft update. Stay
• 16:31
  patched. Stay skeptical. That's the golden rule.
• 16:34
  When you hear a terrifying leak claim or see
• 16:37
  a massive headline designed to induce panic,
• 16:40
  take a breath and look for the verification.
• 16:42
  Assess the actual mechanics. Don't let the noise
• 16:44
  distract you from securing the everyday infrastructure
• 16:46
  that actually runs your life. But, you know,
• 16:48
  connecting all of these threads does leave us
• 16:51
  with a rather profound paradox. Oh. What's that?
• 16:54
  Well, today we learned that AI is pivoting away
• 16:56
  from consumer toys to actively simulate and manage
• 17:00
  our physical reality. Right. And simultaneously,
• 17:02
  we saw that our mundane third -party supply chains,
• 17:06
  the everyday vendors we rely on, are our absolute
• 17:09
  biggest security vulnerabilities. Yeah, the soft
• 17:12
  targets. So it raises a critical question for
• 17:14
  you to think about moving forward. As AI agents
• 17:17
  become advanced enough to seamlessly integrate
• 17:20
  into our complex business networks and manage
• 17:22
  our physical logistics, will these AI agents
• 17:25
  be the ultimate solution to finally securing
• 17:28
  our messy human error prone supply chains? Or
• 17:31
  will turning over our critical infrastructure
• 17:34
  to opaque AI agents become the most dangerous
• 17:36
  third party vulnerability we have ever created?
• 17:38
  Wow. Will the ultimate fix just become the ultimate
• 17:41
  soft target? Exactly. That is a truly fascinating
• 17:44
  and... slightly terrifying thought to leave you
• 17:47
  with. Thank you so much for joining us on this
• 17:49
  deep dive. Stay patched, stay skeptical, and
• 17:51
  we will catch you next time.