Latest / Tech Talks With Kinsoft / Last Week in Tech – Big Tech's "We'll Build It For You" Land Grab, a Cheaper Claude, and a CitrixBleed Sequel

Transcript

Last Week in Tech – Big Tech's "We'll Build It For You" Land Grab, a Cheaper Claude, and a CitrixBleed Sequel · Tech Talks With Kinsoft

• 0:00
  For the past, I don't know, few years maybe,
• 0:01
  you and I have been sold this really beautiful,
• 0:05
  frictionless vision of the future. Right, the
• 0:08
  whole artificial intelligence as magic in a box
• 0:10
  pitch. Exactly. I mean, the pitch is always the
• 0:12
  same, right? You buy the super expensive corporate
• 0:14
  subscription, you plug it into your company's
• 0:16
  systems, and just boom. Instantly, your entire
• 0:19
  business is transformed. Yeah, you're supposed
• 0:21
  to get immediate efficiency, massive profit margins,
• 0:24
  and a workforce that suddenly just has superpowers.
• 0:27
  It's an incredibly seductive idea. Yeah. We all
• 0:30
  want the easy button, especially when you're
• 0:32
  navigating these massive, confusing shifts in
• 0:35
  technology. Oh, absolutely. But, well, when we
• 0:38
  look at the diagnostic landscape from Kinsoft,
• 0:40
  specifically covering the first week of July
• 0:42
  2026, we are presented with a massive reality
• 0:46
  check. A huge one. Yeah. We are looking at an
• 0:48
  industry -wide pivot. The tech giants are moving
• 0:50
  rapidly away from that plug -and -play myth,
• 0:52
  and they are colliding hard with the reality
• 0:55
  of how these tools actually... function or fail
• 0:58
  to function inside a real organization. So today's
• 1:02
  deep dive is all about cutting through that hype.
• 1:04
  We're looking at a stack of fresh reports and
• 1:06
  data to see what actually works, what fails spectacularly,
• 1:10
  and where the real threats lie for you and your
• 1:15
  organization. Because the landscape is shifting
• 1:17
  under our feet as we speak. It really is. We
• 1:21
  are going to explore how major tech giants are
• 1:24
  fundamentally changing how they sell AI, the
• 1:28
  sudden rise of hyper -capable autonomous agents,
• 1:30
  and how these shifts are, well, completely rewriting
• 1:34
  the cybersecurity battlefield. It's a lot to
• 1:36
  cover. Okay, let's unpack this. We have to start
• 1:38
  with the biggest and honestly the most surprising
• 1:40
  theme from the sources. Because the headline
• 1:43
  story this week isn't a new model release. Right.
• 1:46
  No flashy new gadgets. Yeah. It's the fact that
• 1:48
  the biggest AI providers on the planet are changing
• 1:50
  their entire business models right before our
• 1:52
  eyes. The shift is entirely structural. On July
• 1:56
  2nd, Microsoft launched something called the
• 1:58
  Microsoft Frontier Company. And this isn't just
• 2:00
  like a small consulting wing, right? No, not
• 2:03
  at all. This is a $2 .5 billion operating business.
• 2:07
  They staffed it with around 6 ,000 engineers
• 2:10
  and industry specialists. Wow. Yeah. And their
• 2:13
  explicit mandate is to embed directly inside
• 2:16
  customer organizations to build and run AI systems
• 2:19
  right there on site. That is wild. And just two
• 2:23
  days prior to that, so on June 30th, Amazon's
• 2:26
  AWS stood up a $1 billion Ford deployed engineering
• 2:30
  group. Right. The parachute pods. Exactly. The
• 2:34
  sources literally describe them as pods of engineers
• 2:36
  who parachute into a business for 45 days. And
• 2:39
  their whole mission is to just build working
• 2:41
  technology, make sure the client is actually
• 2:43
  self -sufficient, and then extract themselves.
• 2:45
  And we should note, OpenAI and Anthropic launched
• 2:48
  very similar ventures back in May. So every major
• 2:51
  provider has suddenly realized that just handing
• 2:54
  you a software license and wishing you luck isn't
• 2:56
  working anymore? Not at all. And the why behind
• 2:58
  this is staggering. The sources point to some
• 3:00
  MIT research showing that... 95 % of enterprise
• 3:03
  generative AI pilots deliver zero measurable
• 3:06
  impact on the bottom line. 95%. I mean, that
• 3:08
  statistic should make any executive pause. Yeah.
• 3:11
  Zero impact. It indicates that despite the immense
• 3:14
  capability of the underlying technology, nearly
• 3:17
  everyone is failing to convert it into. actual
• 3:21
  measurable business value. Right. So the vendors
• 3:24
  have looked at this massive failure rate and
• 3:26
  realized, hey, the bottleneck isn't the intelligence
• 3:29
  of the model. Because the models work beautifully
• 3:31
  in a vacuum. Exactly. The bottleneck is the messy,
• 3:35
  complicated reality of corporate implementation.
• 3:38
  To put this in perspective for you, imagine going
• 3:41
  to a store and buying a massive state -of -the
• 3:44
  -art IKEA kitchen. Okay, I like this. You've
• 3:47
  got all the flat pack boxes. You've got the glossy
• 3:49
  catalog. You're incredibly excited to start cooking.
• 3:52
  But there's a problem. A huge problem. Ikea looks
• 3:55
  at their customer data and realizes that 95 %
• 3:59
  of the people buying this beautiful kitchen.
• 4:02
  They don't actually know how to use an Allen
• 4:03
  wrench. Right. And their home plumbing doesn't
• 4:06
  match the Swedish metric system. And I don't
• 4:08
  know, their drywall is just crumbling. Structural
• 4:11
  disaster. Exactly. So rather than letting their
• 4:14
  expensive products sit in boxes in your living
• 4:16
  room, gathering dust and risking you calling
• 4:18
  them up and asking for a refund. IKEA realizes
• 4:21
  they have no choice. They have to step in. Yeah,
• 4:24
  they are suddenly forced to dispatch 6 ,000 master
• 4:26
  carpenters directly to your house just to make
• 4:29
  sure the kitchen actually gets built. What's
• 4:31
  fascinating here is how this entirely alters
• 4:33
  the relationship between you and the vendor.
• 4:35
  How so? Well, because corporate data is so deeply
• 4:38
  siloed and legacy workflows are so entrenched,
• 4:41
  change management is just incredibly difficult.
• 4:43
  So these tech giants are essentially taking over
• 4:46
  the implementation phase entirely. Which sounds
• 4:49
  great on paper. It does, but there is a massive
• 4:52
  caveat for you to consider here. Having Microsoft
• 4:55
  or Amazon send their elite engineers to build
• 4:57
  your custom AI solutions, it's fast. It absolutely
• 5:01
  clears that immediate hurdle. But there's a catch.
• 5:04
  A big one. It deeply entrenches your dependence
• 5:07
  on their specific platforms. Oh, I see. You are
• 5:10
  effectively handing them the architectural blueprints
• 5:12
  to your entire operational infrastructure. Exactly.
• 5:16
  It really raises the stakes on the classic build
• 5:19
  versus buy versus partner dilemma. Because if
• 5:21
  you partner to this degree, the lock -in is severe.
• 5:25
  So if your own internal AI pilots have quietly
• 5:28
  gone nowhere, you shouldn't feel bad. You are
• 5:31
  overwhelmingly the norm. You are the 95%. But
• 5:35
  you have to be incredibly clear -eyed about who
• 5:37
  these parachuting engineers actually work for
• 5:40
  at the end of the day. Right, because they aren't
• 5:42
  your employees. No. They are there to integrate
• 5:44
  your company so deeply into their proprietary
• 5:46
  ecosystem that leaving them later on becomes,
• 5:49
  well, financially ruinous. They work for the
• 5:52
  platform's bottom line, not yours. Precisely.
• 5:55
  So this begs a massive question. If enterprise
• 5:57
  implementation at the top level is failing 95
• 6:00
  % of the time and it requires a literal SWAT
• 6:03
  team of engineers to fix, how do everyday businesses
• 6:06
  actually afford to run AI at scale? That is the
• 6:09
  million -dollar question. And the sources point
• 6:11
  to the newly released mid -tier models as the
• 6:13
  missing link here. Okay, mid -tier, like Anthropic's
• 6:16
  new release. Yes. On June 30th, Anthropic released
• 6:20
  Claude Sonnet 5. And they claim it performs very
• 6:24
  close to their top -end Opus model, but at an
• 6:27
  absolute fraction of the cost. We are talking...
• 6:30
  $2 per million input tokens and $10 per million
• 6:33
  output tokens. Right. And the terminology there
• 6:36
  is really important to clarify because the underlying
• 6:38
  mechanics dictate the economics of all this.
• 6:40
  Okay, break that down for us. Think of a token
• 6:42
  roughly as a piece of a word or maybe a syllable.
• 6:45
  And the reason there is such a stark difference
• 6:47
  in price between input and output comes down
• 6:50
  to computational effort. Got it. Input tokens
• 6:53
  are what you feed the AI it is essentially reading.
• 6:56
  It costs Anthropic far less computing power to
• 7:00
  have their model scan a million tokens of your
• 7:02
  internal corporate documents than it does to
• 7:05
  have the AI actively generate or write a million
• 7:08
  new tokens from scratch. Which is the output.
• 7:10
  Exactly. Output takes much more compute. So if
• 7:12
  I'm understanding this, input is like handing
• 7:14
  the AI a textbook and telling it to read it.
• 7:16
  And output is asking the AI to write a completely
• 7:19
  original thesis based on that textbook. That's
• 7:22
  a great analogy. Writing takes way more brainpower,
• 7:24
  so they charge you five times as much for it.
• 7:27
  Precisely the mechanism at play. But the critical
• 7:30
  detail with Sonnet 5 isn't just the drastic drop
• 7:33
  in the price of these tokens. It's the actual
• 7:35
  functionality. What can it do that's different?
• 7:38
  Well, this model is specifically tuned to run
• 7:40
  as an autonomous agent. It isn't just a chat
• 7:43
  interface sitting there waiting for your prompt.
• 7:45
  Right. It can plan multi -step tasks. It can
• 7:48
  navigate a live web browser or operate a command
• 7:51
  line terminal completely on its own to achieve
• 7:53
  a goal you set for it. Okay, here's where it
• 7:56
  gets really interesting, though. And I kind of
• 7:58
  want to push back on the logic here for a second.
• 8:00
  Go for it. We just established that 95 % of corporate
• 8:04
  AI pilots are failing because corporate infrastructure
• 8:07
  and data is a complete mess. Right. So if the
• 8:10
  bottleneck is implementation and messy data,
• 8:13
  does making a cheaper mid -tier model that acts
• 8:16
  autonomously actually solve that implementation
• 8:19
  bottleneck? Or does this just make it cheaper
• 8:21
  for companies to fail at a much higher volume?
• 8:23
  That's a very fair point. But if we connect this
• 8:26
  to the bigger picture, the mid -tier models actually
• 8:30
  solve a massive piece of the unit economics puzzle.
• 8:33
  How so? Well, the headline models, the massive
• 8:36
  expensive flagship AIs from OpenAI or Google,
• 8:40
  they get all the media attention. But the mid
• 8:42
  -tier is where the real daily work actually happens.
• 8:45
  Because it's cheaper. Yes. Think about it this
• 8:48
  way. If you priced up a business automation six
• 8:51
  months ago, say, having an AI independently open,
• 8:54
  review, cross -reference, and process thousands
• 8:57
  of vendor invoices, the numbers likely didn't
• 9:01
  stag up at all. Because using the flagship model
• 9:03
  for that would be too expensive. Exactly. The
• 9:05
  compute cost of having a top -tier model do that
• 9:08
  repetitive background work would completely eat
• 9:10
  your entire profit margin. Right, because you
• 9:12
  can't have an AI doing thousands of mundane,
• 9:14
  everyday tasks if it costs you a dollar every
• 9:17
  single time it clicks a button. Exactly. But
• 9:19
  this new pricing structure completely changes
• 9:22
  the math for your business. Suddenly, processing
• 9:24
  those invoices autonomously costs pennies. It
• 9:27
  actually makes financial sense. Oh, I see. However,
• 9:30
  utilizing an autonomous agent like that changes
• 9:32
  your risk profile entirely. Security -wise. Yes.
• 9:36
  An agent with access to a browser or a terminal
• 9:39
  needs to be treated like a brand new human employee.
• 9:42
  It requires strict guardrails, extensive activity
• 9:46
  logging, and crucially, least privilege access.
• 9:50
  Just to clarify that concept for the listener,
• 9:52
  least privilege means you treat the AI exactly
• 9:55
  like you would, say, a brand new summer intern.
• 9:59
  That's the perfect way to look at it. You give
• 10:00
  the intern the keys to the specific filing cabinet
• 10:03
  they need for today's project, but you absolutely
• 10:05
  do not give them the master passcode to the server
• 10:08
  room so they can accidentally delete the entire
• 10:11
  company database. Yes. If you let an AI loose
• 10:14
  in your terminal with sweeping administrative
• 10:15
  rights, a single hallucination or just a misunderstood
• 10:19
  instruction could result in absolute catastrophe.
• 10:22
  Which is a terrifying thought. But, you know,
• 10:24
  if legitimate businesses are rushing to build
• 10:26
  infrastructure for these cheap agents to automate
• 10:28
  their accounting and HR departments, it tells
• 10:31
  us something massive is shifting underneath the
• 10:33
  surface. The plumbing is being upgraded. Exactly.
• 10:36
  To power millions of new, cheap, autonomous agents,
• 10:40
  an unbelievable amount of capital is quietly
• 10:43
  flowing into the underlying plumbing of the AI
• 10:45
  world. We are seeing massive investments there.
• 10:48
  Yeah. On July 1st, a firm called Together AI,
• 10:51
  which essentially provides infrastructure to
• 10:54
  help businesses run open source AI cheaply, they
• 10:57
  raised $800 million. Wow. Yeah, that puts their
• 11:01
  valuation at an astonishing $8 .3 billion. And
• 11:05
  just purely as a factual financial detail from
• 11:07
  the sources, that funding round was led by the
• 11:10
  venture arm of the Saudi oil giant. That capital
• 11:13
  influx is very much part of a broader trend of
• 11:16
  fortifying the foundation layer. Yeah. Bloomberg
• 11:18
  also recently reported that Meta is building
• 11:20
  a commercial cloud business of its own, internally
• 11:23
  called MetaCompute. Right, with the goal of renting
• 11:25
  out its spare AI computing power to outside customers,
• 11:28
  though we should explicitly clarify that this
• 11:30
  is an unconfirmed press report based on anonymous
• 11:32
  sources. Yes, definitely. We are treating it
• 11:34
  purely as reported, not official. But the trend
• 11:36
  is clear. So what does this all mean for you?
• 11:38
  When I see oil giants and social media monoliths
• 11:42
  pouring billions into server farms and open source
• 11:44
  hosting frameworks, it feels exactly like the
• 11:47
  1849 gold rush. The pickaxes and shovels. Exactly.
• 11:51
  Everyone remembers the miners out in the rivers
• 11:53
  panning for gold, hoping to strike it rich on
• 11:56
  a lucky find. But the people who built the real
• 11:59
  generational fortunes weren't the gold miners.
• 12:02
  No, they weren't. It was the people selling the
• 12:04
  pickaxes, the people manufacturing the durable
• 12:07
  denim jeans, and the people building the railroads
• 12:09
  to transport the gold out of the hills. That
• 12:12
  is incredibly accurate. The models themselves,
• 12:15
  the chatbots, the autonomous agents, they are
• 12:17
  the miners. But the massive capital is flowing
• 12:20
  into the layer beneath the AI you actually touch.
• 12:24
  The railroads. Yes, it's going into the data
• 12:26
  centers, the compute power, and the hosting frameworks.
• 12:29
  And honestly, for you, the listener, this massive
• 12:32
  influx of capital going into the plumbing is
• 12:34
  entirely positive. Because it drives down prices.
• 12:37
  Exactly. It means fierce competition down at
• 12:40
  the foundation layer. And over time, that competition
• 12:43
  results in better choices, more stability, and
• 12:46
  most importantly, much cheaper computing prices
• 12:48
  for anyone trying to run these systems. OK, so
• 12:51
  we've established that these autonomous AI agents
• 12:54
  are getting incredibly cheap. They're highly
• 12:56
  capable and they have this massive bedrock of.
• 13:01
  well -funded infrastructure supporting them.
• 13:03
  Which is great for innovation. That is fantastic
• 13:05
  news if you are trying to automate your supply
• 13:07
  chain. But it terrifies me to think about what
• 13:10
  happens when malicious actors get their hands
• 13:13
  on that exact same cheap computing power. Yes.
• 13:17
  The implications for cybersecurity here are profound.
• 13:20
  We are no longer dealing with theoretical threats.
• 13:22
  The sources detail research from Sysdig, who
• 13:25
  documented a piece of malware they actually characterized
• 13:27
  as the first fully agentic ransomware. And they
• 13:29
  named it Jada Puffer. Right, Jada Puffer. And
• 13:32
  the key distinction here is that this wasn't,
• 13:34
  you know, a human hacker sitting in a dark room
• 13:37
  using Chad GPT to help them write malicious code
• 13:40
  faster. No, not at all. This was an AI agent
• 13:42
  acting end to end completely on its own. That
• 13:45
  is so creepy. It really is. It broke in, it hunted
• 13:48
  through the network for secrets, and it executed
• 13:51
  the extortion autonomously. How did it even get
• 13:53
  in? It achieved this by exploiting an unpatched
• 13:56
  flaw in an open source AI development tool called
• 13:59
  Langflow. And once it breached the perimeter,
• 14:02
  the AI agent methodically harvested cloud keys,
• 14:06
  grabbed database credentials, and successfully
• 14:08
  encrypted over 1 ,300 configuration items in
• 14:12
  a targeted system. But here's the specific detail
• 14:15
  that absolutely blew my mind. Right. During the
• 14:17
  attack, one of the AI's login attempts actually
• 14:20
  failed. It hit a roadblock. Right. It got rejected.
• 14:22
  But instead of stopping or crashing, the AI agent
• 14:25
  diagnosed the problem, rewrote its own approach,
• 14:28
  and had a working fix. running just 31 seconds
• 14:31
  later. This is the mechanism we really have to
• 14:33
  linger on because this machine speed adaptation
• 14:35
  changes the entire security paradigm. 31 seconds.
• 14:39
  Think about it. When a traditional automated
• 14:41
  script hits a failed login, it usually just crashes
• 14:44
  and generates an error report for the human to
• 14:46
  review later. Right, it just gives up. But Jada
• 14:48
  Puffer parsed the server's error response, understood
• 14:51
  logically why it failed. Maybe the API endpoint
• 14:54
  had changed or the payload formatting was just
• 14:57
  slightly off and then it rewrote its own Python
• 15:00
  script to adjust to the new parameter. And just
• 15:02
  executed it again. Exactly. Wait, 31 seconds.
• 15:05
  If a human hacker hit a failed login. and got
• 15:10
  some unexpected error code, they would still
• 15:12
  be taking a sip of their energy drink. They'd
• 15:14
  be cracking their knuckles, opening up a developer
• 15:16
  forum, trying to figure out what went wrong.
• 15:18
  It would take a human 10 minutes just to realize
• 15:21
  they had a typo in their code. It's true. Doesn't
• 15:24
  this mean the skill floor for launching a devastating
• 15:26
  targeted cyber attack just dropped straight into
• 15:29
  the basement? It really does. This raises an
• 15:31
  incredibly important question about the democratization
• 15:33
  of cyber threats. The skill floor has dropped
• 15:36
  to the literal cost of renting an AI. Exactly.
• 15:41
  Sysdig's characterization of this being the first
• 15:44
  fully agentic ransomware might be their specific
• 15:47
  marketing framing, but the core lesson is undeniable.
• 15:50
  These autonomous agents can sweep up easy wins
• 15:53
  at a speed no human could ever match. They are
• 15:56
  constantly scanning the internet for exposed
• 15:59
  developer tools, unpatched software, default
• 16:01
  passwords. And they don't sleep. They don't sleep,
• 16:04
  they don't get distracted, and as we saw, they
• 16:06
  adapt to roadblocks in 31 seconds. So, how do
• 16:09
  you actually defend your business against a hyper
• 16:12
  -fast, autonomous AI hacker that never sleeps?
• 16:15
  You'd think you need some sci -fi defense system.
• 16:18
  Right. You would naturally assume the answer
• 16:20
  is to go out and buy a million -dollar, next
• 16:22
  -generation AI defense shield to fight fire with
• 16:25
  fire. But surprisingly, the sources indicate
• 16:28
  the answer isn't more AI. It's not. No. The answer
• 16:32
  is fixing incredibly boring, traditional vulnerabilities
• 16:35
  and securing the human layer. Boring defenses.
• 16:38
  Exactly. The most advanced technology in the
• 16:40
  world doesn't matter at all if you leave the
• 16:42
  architectural vulnerabilities exposed. Let's
• 16:45
  look at the boring defenses first. On June 30th,
• 16:48
  Citrix released a patch for a critical flaw in
• 16:50
  its Netscaler networking gear. It's tracked as
• 16:53
  CVE -2026 -8. And this is a big deal, right?
• 16:57
  A massive deal. Security experts are comparing
• 17:00
  the potential fallout to Citrix Bleed, which
• 17:03
  caused immense operational damage a couple of
• 17:05
  years ago. This new vulnerability is a pre -authentication
• 17:08
  memory leak bug. Okay, I want to make sure I
• 17:11
  understand how this actually works. A pre -authentication
• 17:13
  memory leak. It's a mouthful. Yeah. So to understand
• 17:17
  this, imagine a bouncer standing at the front
• 17:20
  door of an exclusive VIP club. The bouncer is
• 17:24
  the authentication system. Okay. Normally, you
• 17:27
  hand him your ID, he checks his list, and he
• 17:28
  lets you in. But with this memory leak bug, instead
• 17:32
  of showing an ID, an attacker tricks the bouncer
• 17:35
  into dropping his clipboard. I see where you're
• 17:37
  going with this. And when the bouncer bends down
• 17:39
  to pick it up, he accidentally blurts out everything
• 17:42
  he is currently thinking about, which happened
• 17:44
  to include the private VIP list. That is an excellent
• 17:47
  way to visualize it. When the device spills chunks
• 17:49
  of its own active memory, an attacker can just
• 17:52
  sift through that raw data and find active session
• 17:55
  tokens. The VIP list. Yes. Those session tokens
• 17:58
  are the VIP list. Once the attacker extracts
• 18:02
  a valid session token, they don't even need to
• 18:05
  show an ID or a password. They just walk right
• 18:07
  in. They bypass the login screen entirely. They
• 18:10
  bypass your multi -factor authentication, stepping
• 18:13
  right into the network as a fully authorized
• 18:15
  user. Wow. And security researchers saw this
• 18:18
  specific net scaler vulnerability being exploited
• 18:21
  in the wild within 24 hours of the patch landing.
• 18:24
  They move fast. So if you run this gear facing
• 18:27
  the open Internet, the source makes it clear.
• 18:29
  This is a patch it today emergency, not a wait
• 18:33
  for the maintenance window next month kind of
• 18:34
  thing. And that is exactly what we mean by a
• 18:36
  boring defense. Diligent patch management isn't
• 18:39
  glamorous. No one is throwing a party for patching
• 18:42
  a server. Right. It doesn't make for an exciting
• 18:44
  keynote presentation. But it is the primary barrier
• 18:47
  standing between your network and an automated
• 18:49
  AI agent looking for a fast, easy win. But the
• 18:52
  technology is only half the battle here, which
• 18:55
  brings us to the human element. Always the weakest
• 18:57
  link. Yeah. On July 1st, the U .S. Department
• 19:01
  of Justice announced the extradition of a 19
• 19:03
  -year -old dual U .S. Esconian citizen from Finland.
• 19:07
  Right. The scattered spider case. Exactly. He
• 19:09
  is an alleged member of the notorious scattered
• 19:12
  spider hacking crew, and he's accused of helping
• 19:15
  breach a luxury retailer to demand about $8 million
• 19:18
  in cryptocurrency. And we must note, these are
• 19:21
  allegations not yet proven in court. Of course.
• 19:23
  But the tactic itself is what we really need
• 19:27
  to focus on. Because Scattered Spider's main
• 19:29
  weapon isn't high -tech machine speed AI. No,
• 19:32
  it's not. It is just old -fashioned social engineering.
• 19:36
  They don't write complex code to break into the
• 19:38
  system. They manipulate humans to talk their
• 19:40
  way in. It's so simple. It is. They locate a
• 19:43
  target, gather some basic information, and then
• 19:45
  they literally just call the human IT help desk.
• 19:48
  On the phone. On the phone. They pretend to be
• 19:51
  a legitimate employee who has, you know, been
• 19:53
  locked out of their account. And they politely
• 19:56
  ask the help desk to reset their access. When
• 19:58
  you put these two distinct threats side by side,
• 20:00
  it's just mind boggling. What do you mean? It's
• 20:02
  like installing a million dollar retinal scanner
• 20:05
  on the front door of your business, but leaving
• 20:07
  the back door completely off its hinges, which
• 20:10
  is ignoring the Netscaler flaw. Right. Or worse,
• 20:13
  having a thief just call the security guard on
• 20:15
  the phone. say they lost their badge, and smoothly
• 20:18
  talk the guard into handing over the master keys,
• 20:21
  which is the scattered spider tactic. That's
• 20:24
  a perfect summary. The best defenses against
• 20:26
  both machine speed AI attacks and human social
• 20:29
  engineers rely on fundamental IT hygiene. The
• 20:33
  basics. Yes. You must patch your internet -facing
• 20:37
  systems immediately. You must kill default credentials.
• 20:41
  You cannot expose your raw development tools
• 20:43
  to the open internet. And what about the human
• 20:45
  side? Well, when it comes to defending against
• 20:48
  humans, the most advanced firewall in the world
• 20:50
  is useless if the human protocol is weak. You
• 20:53
  have to empower your help desk. They need to
• 20:55
  be extensively trained and, frankly, given the
• 20:58
  structural authority to rigorously verify someone's
• 21:01
  identity before they ever reset a password or
• 21:04
  hand over network access. Even if the person
• 21:06
  on the phone is yelling at them. Especially then.
• 21:08
  Man. We have covered a massive amount of ground
• 21:11
  in this deep dive. Let's recap the journey really
• 21:13
  quick. Sounds good. We started with the startling
• 21:15
  reality that 95 % of corporate AI pilots are
• 21:19
  failing to deliver a measurable return, which
• 21:22
  is forcing tech giants like Microsoft and Amazon
• 21:24
  to literally send armies of engineers into businesses
• 21:27
  just to make the technology function within messy
• 21:31
  corporate data silos. The IKEA carpenters. Exactly.
• 21:34
  Then we saw the rise of cheap mid -tier... AI
• 21:37
  models like CloudSonic 5, making autonomous agents
• 21:40
  financially viable for everyday business automations.
• 21:43
  Pennies per task. Right. But we also saw the
• 21:46
  dark side of that cheap computing power, with
• 21:48
  the rise of Jada Puffer and autonomous AI agents
• 21:50
  capable of hacking a system, diagnosing a roadblock,
• 21:54
  and rewriting their own code in just 31 seconds.
• 21:57
  Still hard to believe. It is. And finally, we
• 21:59
  learned that the best defense against all of
• 22:01
  this isn't necessarily more AI. It is diligently
• 22:04
  patching your systems today and ensuring your
• 22:06
  help desk can't be fast -talked by a 19 -year
• 22:09
  -old hacker. It is a rapidly evolving, highly
• 22:12
  complex landscape, but the fundamentals of how
• 22:15
  we protect our organizations remain surprisingly
• 22:18
  constant. They really do. However, I want to
• 22:21
  leave you with a final thought to ponder, something
• 22:23
  that connects all the diverse dots we've discussed
• 22:26
  today. I'm ready. Lay it on us. Today, we talked
• 22:28
  about AI agents acting autonomously to hack systems.
• 22:32
  And we talked about human hackers using social
• 22:34
  engineering to trick human help desks. Right.
• 22:37
  Two separate threats. But look at where these
• 22:39
  two distinct trend lines intersect. Okay. As
• 22:42
  AI agents like CloudSonic 5 get incredibly cheap,
• 22:45
  highly capable, and fully autonomous. What happens
• 22:49
  when the entity calling your help desk to social
• 22:52
  engineer a password reset isn't a 19 -year -old
• 22:54
  kid reading a script? Wait. What if it's an autonomous
• 22:57
  conversational AI agent that has been trained
• 23:00
  to sound exactly like your CEO? Oh, wow. That
• 23:03
  is terrifying. How will your human employees
• 23:06
  verify if the frantic voice on the other end
• 23:08
  of the phone demanding urgent access to the network
• 23:11
  is actually a co -worker in distress or just
• 23:13
  a 31 -second script running on a mid -tier model?
• 23:16
  They wouldn't know. That is the intersection
• 23:18
  we were rushing toward, and it is a reality businesses
• 23:20
  need to prepare for today. That brings us right
• 23:23
  back to where we started. We all wanted the plug
• 23:26
  -and -play magic box that would make everything
• 23:28
  easier. And we didn't get it. No. What we got
• 23:31
  was a landscape where you have to meticulously
• 23:34
  guard your internal data, patch your systems
• 23:37
  daily, and prepare for a future where you can't
• 23:39
  even trust the voice on the other end of the
• 23:41
  phone. It's a brand new world. It really is.
• 23:44
  Yeah. And the easy button. definitely does not
• 23:46
  exist. Thank you for joining us on this deep
• 23:48
  dive. Stay curious, stay patched, and keep exploring.