Latest / Tech Talks With Kinsoft / Last Week in Tech – Microsoft's Agents Go GA, the Canvas Mega-Breach Escalates, and DigiCert's Cert Compromise

Transcript

Last Week in Tech – Microsoft's Agents Go GA, the Canvas Mega-Breach Escalates, and DigiCert's Cert Compromise · Tech Talks With Kinsoft

• 0:00
  Imagine building like the ultimate completely
• 0:02
  impenetrable fortress. I am talking titanium
• 0:05
  walls, laser grids, biometric scanners that check
• 0:10
  your retina and your fingerprints. The whole
• 0:12
  nine yards. Right. The whole nine yards. You
• 0:14
  spend billions on the architecture and it is
• 0:16
  mathematically perfect. And then on day one,
• 0:19
  the security guard stationed at the front desk
• 0:22
  gets a phone call. from someone pretending to
• 0:25
  be, I don't know, the fire inspector. And the
• 0:27
  guard just casually props the titanium backdoor
• 0:30
  open with a brick. Wow. Yeah, it completely invalidates
• 0:33
  the billion -dollar laser grid, right? Because
• 0:35
  the technology worked flawlessly, but the overall
• 0:38
  system still suffered a, well, a catastrophic
• 0:41
  failure just because of that human element. Welcome
• 0:44
  to the Deep Dive. Today, we are looking at a
• 0:47
  recent, incredibly revealing snapshot from the
• 0:49
  tech landscape. It's the Tech Talks with Kinsoft
• 0:52
  update from May 11th, 2026. A really fascinating
• 0:56
  brief. It is, because this single document captures
• 1:00
  this terrifying paradox about where we are right
• 1:04
  now. You know, we are building these mathematically
• 1:06
  perfect digital fortresses, but we are just leaving
• 1:10
  those titanium doors. propped open everywhere.
• 1:13
  Yeah, that's exactly it. On one hand, our technology
• 1:15
  is becoming unbelievably autonomous and capable.
• 1:18
  But then on the other hand, our biggest vulnerabilities
• 1:20
  are still fundamentally human. It really is kind
• 1:25
  of the defining tension of this era we're in.
• 1:28
  When we analyze this Kinsoft source material,
• 1:30
  two major interconnected themes emerge for you
• 1:34
  to really think about. Right. First, we have
• 1:36
  the official mainstream arrival of autonomous
• 1:38
  AI agents. And then running totally parallel
• 1:41
  to that massive leap forward, we have a series
• 1:44
  of just catastrophic cyber breaches. Huge ones.
• 1:47
  Yeah. And the narrative thread tying this all
• 1:49
  together is that as we hand over more power to
• 1:52
  technology to act on our behalf, you know, the
• 1:54
  human gatekeepers managing these systems are
• 1:56
  increasingly becoming our absolute weakest link.
• 1:59
  Okay, let's unpack this because we have to start
• 2:01
  with a big announcement from Microsoft's recent
• 2:02
  build conference. Oh, absolutely. According to
• 2:04
  the Kinsoft update, Microsoft has officially
• 2:07
  made the, quote, agent era a reality. They announced
• 2:11
  that their Agent 365 platform, along with this
• 2:14
  entirely new agent framework and a whole stack
• 2:17
  of in -house AI models, has moved to general
• 2:20
  availability. Right. General availability is
• 2:22
  the key term there. Yeah. Plus they rolling out
• 2:25
  a new unified intelligence layer across all the
• 2:28
  developer tools. So we are officially out of
• 2:31
  the experimental sandbox here. But, you know,
• 2:33
  what does that actually mean for you, the listener?
• 2:35
  We hear words like framework and layer. all the
• 2:38
  time in tech we do but if we connect this to
• 2:41
  the bigger picture the shift to general availability
• 2:43
  is a massive milestone this isn't uh beta testing
• 2:47
  for a few select tech enthusiasts in a basement
• 2:49
  anymore right this is a fully supported enterprise
• 2:52
  -grade product rolling out to the global mainstream
• 2:54
  and the transition here the really important
• 2:56
  part is moving from conversational ai to agentic
• 3:00
  ai okay break that down a bit well For the last
• 3:02
  few years, we have basically been talking to
• 3:04
  incredibly smart chatbots, right? You ask a question,
• 3:08
  it generates some text. But an AI agent is an
• 3:12
  entity designed to take real independent actions
• 3:15
  inside your actual software ecosystem. So rather
• 3:18
  than just like generating a draft of an email
• 3:21
  that I have to manually copy and paste, the agent
• 3:24
  actually reads my inbox, decides an email needs
• 3:27
  to be sent, writes it, and literally hits the
• 3:29
  send button for me. that is the practical application
• 3:31
  yes it's doing the work yeah and to understand
• 3:35
  how that actually functions we have to look at
• 3:38
  that unified intelligence layer you mentioned
• 3:40
  earlier okay think of previous ai as like a brain
• 3:43
  trapped in a jar a brain in a drawer right it
• 3:46
  was really smart but it couldn't touch anything
• 3:47
  yeah this new unified intelligence layer acts
• 3:50
  like a nervous system connecting all your different
• 3:52
  applications it provides the digital roads for
• 3:55
  the agent to travel between your word processor
• 3:57
  your email client your corporate database. And
• 4:01
  the framework itself gives the AI the hands to
• 4:04
  actually push buttons and move files across those
• 4:07
  different software environments. It is honestly
• 4:09
  like hiring a brand new, incredibly fast, hyper
• 4:13
  -efficient digital employee. You bring them on
• 4:15
  board and suddenly you have this massive workforce
• 4:18
  multiplier. Oh, massive. But wait, I have to
• 4:21
  push back here for a second. Think about it like
• 4:23
  corporate credit cards. Okay. If you give a corporate
• 4:26
  card to a human employee, you have an HR file.
• 4:29
  You've done a background check. You have a manager
• 4:31
  actively watching their spending. Hopefully,
• 4:32
  yeah. Hopefully. But if we are handing out digital
• 4:36
  ID badges to these AI agents, you know, giving
• 4:40
  them access to our sensitive corporate environments,
• 4:43
  how do we actually govern a machine's identity
• 4:46
  and its access rights? Oh, the million dollar
• 4:48
  question. Right. Because imagine handing out
• 4:50
  10 ,000 corporate cards in a single second to
• 4:53
  invisible digital workers. Yeah. How do you even
• 4:56
  begin to track who's doing what? You've hit on
• 4:58
  the exact vulnerability the Kinsoft update explicitly
• 5:01
  warns about. Every single agent you deploy is
• 5:05
  a new non -human identity with access rights
• 5:08
  that absolutely must be governed. It's just a
• 5:10
  massive headache waiting to happen. It is. When
• 5:13
  an AI agent logs into a database to pull records,
• 5:16
  the system needs to know if that specific agent
• 5:18
  is authorized. If an attacker manages to hijack
• 5:22
  an agent or, say, trick an agent through a malicious
• 5:26
  prompt. Oh, wow. The attacker inherits all the
• 5:29
  machine's permissions. We are rapidly expanding
• 5:32
  our attack surface by creating thousands of these
• 5:35
  non -human identities that basically hold the
• 5:37
  keys to our digital kingdoms. Which means the
• 5:39
  platforms managing all these identities and all
• 5:41
  this access need to be incredibly. secure, like
• 5:44
  flawlessly secure. And that introduces a major
• 5:46
  structural problem. Right. Because giving AI
• 5:49
  new identities and deep platform access requires
• 5:52
  those underlying platforms to be flawless. But
• 5:55
  the reality of our current infrastructure is,
• 5:57
  well, it's anything but flawless. Which brings
• 5:59
  us to the architecture of the cloud itself. That
• 6:01
  feels like the perfect bridge to the next massive
• 6:04
  story in the Kinsoft update. Since we're giving
• 6:07
  these agents the keys to our cloud platforms,
• 6:10
  we really need to look at how fragile those platforms
• 6:12
  actually are right now. They're more fragile
• 6:14
  than people think. Yeah, and the nightmare scenario
• 6:17
  unfolded recently at Instructure. For those who
• 6:20
  don't know, Instructure runs Canvas, which is
• 6:22
  this massive learning management system used
• 6:25
  by countless schools and universities. Almost
• 6:27
  everyone in education uses it. Right. Well, the
• 6:30
  threat actor, a crew known as Shiny Hunters,
• 6:32
  managed to breach the system. And they didn't
• 6:35
  just quietly steal data. No, it was very loud.
• 6:38
  They actively defaced the Canvas login portals
• 6:41
  of roughly 330 different institutions. They stole
• 6:45
  data belonging to hundreds of millions of users.
• 6:48
  And the situation was so dire that a ransom deal
• 6:51
  was eventually struck. It's just staggering.
• 6:54
  This incident is being called the largest education
• 6:56
  sector breach on record. Unbelievable. And to
• 6:59
  really understand the sheer scale of the devastation
• 7:01
  here, we have to examine how these services are
• 7:04
  actually delivered. Canvas operates as a massive
• 7:07
  SaaS platform, software as a service. Right.
• 7:10
  SaaS. The central lesson here from Kinsoft is
• 7:13
  the terrifying scale of vendor breaches. Break
• 7:17
  down that scale for us a bit. Like, why is a
• 7:19
  SaaS breach fundamentally different from, say,
• 7:21
  a traditional network hack from 10 years ago?
• 7:24
  Well, in the past, if a university wanted a digital
• 7:27
  portal for its students, they bought their own
• 7:29
  physical servers, they installed them in a dusty
• 7:31
  basement on campus, and they ran the software
• 7:34
  locally. Okay. An attacker would have to breach
• 7:36
  University A's firewalls and then start completely
• 7:39
  from scratch to separately breach University
• 7:41
  B's firewalls. It was a lot of manual work for
• 7:44
  the hacker. Exactly. But with a multi -tenant
• 7:47
  sauce model, all those universities outsource
• 7:49
  their infrastructure to a single centralized
• 7:51
  vendor. Instructure is the vendor. It is essentially
• 7:55
  a giant apartment building where everyone shares
• 7:57
  the exact same foundation and the same plumbing.
• 7:59
  So the attacker doesn't need to break into 330
• 8:02
  separate university servers. They just need to
• 8:04
  find like one vulnerability in the master key
• 8:08
  system of the apartment building. You got it.
• 8:10
  And once that single central hub is compromised,
• 8:14
  the disaster cascades instantly to every single
• 8:17
  tenant. The economies of scale that make sauce
• 8:19
  so affordable and so efficient. Which is why
• 8:22
  everyone uses it. Right. Those same economies
• 8:24
  of scale also create massive single points of
• 8:27
  failure. It's kind of terrifying when you think
• 8:29
  about it. You drop one massive cloud -shaped
• 8:32
  basket. And 330 institutions lose hundreds of
• 8:36
  millions of user records. I mean, grades, personal
• 8:39
  information, financial data, all at the exact
• 8:41
  same time. And think about the ransom negotiation.
• 8:44
  Oh, God. How does a single vendor negotiate a
• 8:46
  ransom when the data belongs to hundreds of independent
• 8:49
  universities? The legal and logistical nightmare
• 8:52
  is just staggering. Seriously. So what does this
• 8:55
  all mean for the listener? I mean, whether you
• 8:57
  are a student, a professional, or running a business,
• 9:00
  we all rely on massive... platforms every single
• 9:04
  day. If the platforms holding our most sensitive
• 9:06
  data can be breached and held for ransom by crews
• 9:09
  like Shiny Hunters, is any data actually safe?
• 9:13
  Like, is this just the cost of doing business
• 9:16
  in 2026? It forces a very sobering reality check
• 9:19
  about vendor risk. You can have the most draconian,
• 9:23
  hyper -secure password policies in your own organization.
• 9:26
  But if your vendor's security fails, your data
• 9:29
  is completely exposed. Wow. It completely shifts
• 9:32
  the burden of trust away from your internal IT
• 9:35
  team and straight on to a third party. But let's
• 9:38
  play devil's advocate for a second. Let's say
• 9:39
  a university gets spooked by the campus breach
• 9:41
  and decides to pull all their data out of the
• 9:44
  cloud. They move everything back to those isolated
• 9:46
  servers in the campus basement. They are completely
• 9:48
  safe now, right? They are not. Not at all. No,
• 9:50
  because even offline, isolated servers still
• 9:52
  need to download software updates, right? To
• 9:54
  patch known bugs. And that exposes a completely
• 9:57
  different... much deeper vulnerability. Oh boy.
• 10:00
  Breaking into a software platform like Canvas
• 10:02
  is one thing, but breaking the system that tells
• 10:04
  the internet what software to trust in the first
• 10:07
  place, that is a foundational crisis. Which perfectly
• 10:10
  leads us to a story that might seem, I don't
• 10:13
  know, a little abstract at first glance, but
• 10:15
  is arguably the most unsettling part of this
• 10:17
  entire Kinsoft update. Definitely. We're talking
• 10:20
  about the breach at DigiCert. According to the
• 10:22
  source, attackers successfully manipulated DigiCert
• 10:25
  into issuing fraudulent code signing certificates.
• 10:28
  And then the attackers used those fake certificates
• 10:31
  to digitally sign their malware, making the malicious
• 10:34
  software look entirely legitimate to literally
• 10:37
  anyone who downloaded it. What's fascinating
• 10:40
  here is the underlying mechanism of Internet
• 10:42
  security that this attack actually exploited.
• 10:45
  We really need to talk about the concept of a
• 10:47
  trust anchor. A trust anchor. Yeah. Whenever
• 10:49
  your computer downloads a piece of software or,
• 10:52
  say, an app, your operating system does a quick
• 10:54
  background check before allowing it to install.
• 10:56
  Right. It asks, who created this file and has
• 10:59
  the code been tampered with since it left the
• 11:01
  developer's computer? It's looking for that little
• 11:03
  pop -up that says, like, verified publisher.
• 11:06
  Exactly. And the way it verifies that publisher
• 11:09
  is through public key cryptography. Certificate
• 11:12
  authorities, like DigiCert, are the organizations
• 11:15
  trusted by Apple, Microsoft, Google to issue
• 11:18
  those cryptographic certificates. Okay, so they're
• 11:21
  the authorities. They are the digital passport
• 11:23
  agencies of the internet. When DigiCert signs
• 11:26
  a certificate for a software developer, it is
• 11:28
  mathematically vouching for them. Gotcha. The
• 11:31
  operating system sees the DigiCert signature
• 11:33
  and applies a fundamental hard -coded rule. This
• 11:37
  file is digitally signed by a trusted anchor.
• 11:39
  Therefore, it is safe to run. So it's like a
• 11:42
  master counterfeiter. Right. But instead of just
• 11:44
  printing fake money in a basement and hoping
• 11:46
  a cashier doesn't look too closely. Right. The
• 11:48
  counterfeiter actually manages to trick the official
• 11:50
  government mint into stamping their fake money
• 11:53
  with a real mathematically perfect watermark.
• 11:56
  The money itself is fundamentally counterfeit,
• 11:59
  but the security features. are 100 % genuine.
• 12:01
  That captures the severity perfectly. By issuing
• 12:05
  those fraudulent certificates, DigiCert unknowingly
• 12:08
  gave malware a VIP pass right past the bouncers
• 12:12
  of every single computer security system. The
• 12:16
  fundamental guarantee of, is this digitally signed,
• 12:18
  was just completely broken. Wait, wait, you're
• 12:20
  losing me here. I have to push back on this.
• 12:22
  Sure. We are talking about DigiCert. A company
• 12:26
  whose entire existence is based on mathematically
• 12:29
  perfect billion dollar cryptography. Yes. We're
• 12:33
  talking about prime numbers so large that it
• 12:35
  would take a supercomputer like a million years
• 12:38
  to crack them. And you're telling me the bad
• 12:40
  guys bypassed all that math just by calling customer
• 12:44
  service and asking nicely. I know. How does a
• 12:46
  help desk worker even have the authority to issue
• 12:48
  a cryptographic certificate? It sounds totally
• 12:50
  absurd, but it highlights a massive blind spot
• 12:53
  in how we structure security. Customer support
• 12:55
  portals have administrative overrides. Oh, of
• 12:58
  course they do. Right. They're designed to help
• 13:00
  legitimate customers who lose their keys or need
• 13:02
  expedited service. The attackers didn't use some
• 13:05
  brilliant zero -day hacking code to break into
• 13:07
  DigiSearch cryptographic vaults. No code at all.
• 13:10
  No. The Kinsoft update explicitly states they
• 13:13
  used social engineering through the support portal.
• 13:16
  They just manipulated a human being who had authorized
• 13:19
  access to the tools that generate those certificates.
• 13:22
  So the foundational trust anchors of the entire
• 13:24
  Internet really were defeated by a smooth talker
• 13:28
  on the help desk line. They were. Honestly, it
• 13:32
  makes perfect sense from the attacker's perspective.
• 13:34
  How so? We spend billions analyzing code, patching
• 13:38
  servers, and configuring firewalls. But attackers
• 13:41
  realize that the path of least resistance is
• 13:43
  almost always human psychology. Yeah, why break
• 13:46
  the math when you can break the person? Exactly.
• 13:48
  Why spend months trying to crack an encryption
• 13:51
  algorithm when you can just convince an employee
• 13:53
  that you are a frantic developer facing a massive
• 13:57
  deadline who needs a certificate reissued immediately?
• 14:00
  Wow. Social engineering relies on manipulation,
• 14:03
  you know, creating a false sense of urgency,
• 14:05
  leveraging fake authority, or just exploiting
• 14:07
  natural human empathy to trick a person into
• 14:09
  bypassing all the rules. Which brings us to the
• 14:12
  ultimate takeaway from the Kinsoff brief, and
• 14:14
  the realization that the human element is sort
• 14:16
  of the universal bypass here. And this wasn't
• 14:19
  just an isolated incident at DigiCert. Not at
• 14:22
  all. The update heavily emphasizes recent breaches
• 14:24
  at the global fashion retailer Zara and at an
• 14:28
  NVIDIA cloud gaming partner. And the common thread,
• 14:31
  Kinzov calls social engineering, quote, The week's
• 14:34
  quiet villain. A great phrase. Let's dive into
• 14:37
  how these specific breaches happened. Because
• 14:39
  Zara and NVIDIA are massive companies with huge
• 14:42
  security budgets. They have multi -factor authentication,
• 14:46
  identity protocols that works. How does a smooth
• 14:48
  talker bypass all of that? The tactics are incredibly
• 14:51
  sophisticated, yet entirely non -technical. In
• 14:54
  scenarios like the NVIDIA Cloud Gaming Hartner
• 14:57
  breach, attackers often utilize a technique called
• 15:00
  MFA fatigue combined with IT impersonation. MFA
• 15:03
  fatigue. Yeah. So they acquire a target's username
• 15:06
  and password, often from an older, completely
• 15:08
  unrelated data leak. They try to log in, which
• 15:11
  triggers a multi -factor authentication request
• 15:13
  to the real employee's phone. The classic press
• 15:16
  one to approve this login notification. Yes.
• 15:19
  And the attacker will spam that notification
• 15:21
  to the employee's phone 50 times in the middle
• 15:24
  of the night. Oh, that's incredibly annoying.
• 15:26
  Right. The employee is annoyed, confused, and
• 15:29
  just wants their phone to stop buzzing at two
• 15:31
  in the morning. But they usually don't click
• 15:33
  approve. Good. But then the next morning, the
• 15:36
  attacker calls the employee, spoofing the phone.
• 15:39
  number so it looks like the internal IT help
• 15:42
  desk oh no the attacker says hey we noticed some
• 15:46
  weird login activity on your account last night
• 15:49
  we are trying to secure it but I need you to
• 15:51
  approve the notification I'm sending to your
• 15:53
  phone right now to verify your identity oh wow
• 15:56
  so the employee thinks they're talking to the
• 15:59
  good guys right I think clicking approve is securing
• 16:02
  their account when in reality they're letting
• 16:04
  the attacker straight into the network exactly
• 16:06
  and similar tactics are used against Exploiting
• 16:23
  that authority. Yes. They bypass the technical
• 16:26
  identity protocols by manipulating the human
• 16:28
  desire to be helpful or, honestly, just the human
• 16:32
  fear of getting in trouble with the boss. It
• 16:34
  is the brick. propping open the titanium door
• 16:37
  all over again. It really is. The laser grid
• 16:39
  works perfectly, but the security guard was tricked
• 16:42
  into turning it off. You know, if we synthesize
• 16:44
  everything we've discussed today, a very stark
• 16:47
  picture emerges. We are deploying incredibly
• 16:50
  advanced autonomous AI agents through platforms
• 16:54
  like Microsoft's Agent 365, giving them their
• 16:58
  own digital corporate credit cards to roam our
• 17:00
  networks. We are centralizing hundreds of millions
• 17:03
  of user records in massive SaaS infrastructures
• 17:06
  like Canvas, creating incredibly lucrative targets.
• 17:09
  The technological stakes have never been higher.
• 17:11
  Never. Yet despite all of this advanced technology,
• 17:14
  attackers are succeeding through completely low
• 17:17
  -tech methods. Right. using convincing phone
• 17:19
  calls, manipulated emails, and support portal
• 17:22
  trickery to bypass the strongest digital defenses
• 17:24
  we can possibly build. The technology isn't failing.
• 17:27
  The human gatekeepers are being manipulated into
• 17:29
  opening the doors. And the Kinsoft update ends
• 17:32
  with a very direct piece of advice that connects
• 17:35
  this right back to you, the listener. Their sign
• 17:38
  -off is stay patched, stay skeptical. Good advice.
• 17:42
  Because whether you were running a massive corporate
• 17:44
  IT department or you're just sitting at home
• 17:46
  trying to figure out if that text message from
• 17:48
  your bank is a phishing scam, the human element
• 17:51
  is the final line of defense. It absolutely is.
• 17:54
  Critical thinking is a required security control
• 17:57
  now. Knowledge is most valuable when it is understood
• 18:00
  and applied. We can build the most robust AI
• 18:03
  frameworks and secure our sauce environments
• 18:05
  with mathematical precision. But if we don't
• 18:08
  train ourselves to view our digital interactions
• 18:10
  with a healthy dose of that recommended skepticism,
• 18:12
  the technical defenses simply do not matter.
• 18:15
  They're useless. Pretty much. The attackers will
• 18:17
  always target the easiest way in. And right now,
• 18:20
  that is us. So to quickly recap this incredibly
• 18:23
  dense snapshot of our tech landscape, we are
• 18:26
  officially entering the era of. autonomous AI
• 18:28
  agents taking real actions inside our software
• 18:31
  nervous systems. At the same time, we are witnessing
• 18:34
  the terrifying single point of failure vulnerability
• 18:37
  of massive SaaS platforms with breaches affecting
• 18:40
  hundreds of millions of people at once. And underneath
• 18:44
  it all, the very cryptographic trust anchors
• 18:46
  of the internet are being bypassed, not with
• 18:49
  complex code, but with simple human conversation
• 18:52
  and psychological manipulation. It is a wild
• 18:55
  time to be navigating the digital world. It certainly
• 18:58
  is. The landscape requires a constant active
• 19:00
  reevaluation of where our true vulnerabilities
• 19:02
  actually lie. And as we wrap up today's deep
• 19:05
  dive, I want to leave you with a final lingering
• 19:07
  thought to mull over. We spent a lot of time
• 19:09
  talking about how easy it is to socially engineer
• 19:11
  a human being like the support staff at DigiCert
• 19:14
  or the employees at an NVIDIA partner into making
• 19:18
  a terrible mistake. But think about those new
• 19:20
  autonomous AI agents Microsoft is rolling out.
• 19:24
  These agents have their own digital identities,
• 19:26
  their own access rights, and the ability to interact
• 19:29
  with a broader digital ecosystem on our behalf.
• 19:32
  What happens when those AI agents become the
• 19:34
  targets of these exact same social engineering
• 19:37
  tricks? We know a smooth talker can manipulate
• 19:39
  a tired human help desk worker by creating a
• 19:42
  false sense of urgency. But as these AI models
• 19:45
  read our emails and interact with outside data,
• 19:48
  I mean, can you fast talk an AI into handing
• 19:51
  over the keys to the kingdom? That is a terrifying
• 19:53
  thought. That is something to think about the
• 19:55
  next time you authorize a digital agent to manage
• 19:57
  your inbox. Thank you for joining us on this
• 19:59
  deep dive. Stay skeptical out there, and we will
• 20:02
  catch you next time.