Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

Tech Talks With Kinsoft

Tech Talks With Kinsoft artwork

Last Week in Tech – Copilot's Billing Backlash, SoftBank's €75B France Bet, and the Klue OAuth Supply-Chain Hit

Jun 7, 202621 minAudio

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! Your Monday catch-up, with a security lens. This week: GitHub Copilot switches to token-based billing and developers revolt over surprise bills, Microsoft shows off a cheaper homegrown coding model at Build, and SoftBank commits up to €75B to AI data centres in France - Europe's biggest such bet. On the breach desk: an OAuth attack on sales-intelligence firm Klue cascades into Salesforce data theft across HackerOne, Gong, OneTrust, Tanium and Huntress; Nintendo is hit by ransomware; and Oxford's careers platform is breached. The OAuth supply-chain pattern is the one to watch. Using connected SaaS apps and OAuth integrations? Visit www.kinsoft.com.au to talk through your security and IT needs. Sources: BuildFastWithAI; Medium (David Akpovi); imfounder; SharkStriker; The Hacker News.

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

So today, an AI coding assistant just build a small startup like $800 just for processing math. Yeah, casual $800. Right. While at the same time, a 75 billion euro power grid expansion is literally being mapped out in France just to keep that kind of math running. It's massive. And on top of all that, Wikipedia volunteers are actually striking because human workers are being laid off to pay for it all. So, welcome to the Deep Dive. Glad to be here. We are so glad to have

you with us today. We've got a stack of sources to go through, anchored by a really revealing briefing from Kinsoft Tech Talks. It's dated June 8th, 2026. Right. The Kinsoft briefing. Yeah. And our mission today for this deep dive is to look at the severe structural reality check that is hitting the tech industry right now because we are definitely moving past that honeymoon phase of the seamless AI -powered upgrade. Oh, the honeymoon is definitely over. Totally over. We're entering this era of steep, multifaceted

consequences. So we're going to track the shock. of metered ai billing models yeah the actual physical limits of global infrastructure and and this is a big one how our rush to integrate all these powerful tools has created just catastrophic blind spots in corporate security it's a really critical set of sources to synthesize i mean because it forces us to look at the hidden causality of the tools we use every single day right because we tend to treat software as you know infinite and weightless like it's just magic in the cloud

exactly magic but But what this briefing exposes is that we are slamming into hard physical and economic limits. The entire tech ecosystem is currently scrambling to re -architect itself before those limits completely break the current business models. Yeah. So let's start with that economic limit, which really hits right at the developer's wallet. Where it hurts. Exactly. The Kinsoft briefing details what happened on June 1st. So GitHub changed the billing structure for its AI coding assistant, Kotilot, for a really long time. This was just a flat monthly subscription.

Twenty nine dollars. Nice and easy. Super predictable. Right. You put it on the corporate card and you just forget about it. Yeah. But then they flipped it to a token based pay as you use billing model. Yeah. And I mean, the reaction was explosive. Oh, people were furious. Yeah, because developers were opening their dashboards expecting that standard twenty nine dollar charge. And instead they're seeing. bills for hundreds of dollars a month. Yeah, which fundamentally shatters the financial predictability that, honestly, the entire software as a service model has been built

on for the last 20 years. Right. Okay, let's unpack this. Because, you know, if you're listening, think about it like this. It's like going to your favorite all -you -can -eat buffet. Okay, I like buffets. Right. You pay your 20 bucks at the door, you eat as much as you want. But suddenly management decides to change the rules and now they're charging you per individual bite you take. That is a fast way to ruin a dinner. Exactly. How are everyday businesses supposed to budget for these tools when AI powered suddenly means it's metered? Well, when you shift from

a flat sauce license to metered token billing, you are exposing the end user to the raw fluctuating cost of compute. But I want to push back on that justification a little bit because I get it. AI inference is expensive. Sure. Right. But we've had cloud computing and massive database lookups for. decades, right? And we didn't suddenly start billing users per keystroke. So why does generating a line of code in an IDE suddenly require like a taxi meter? That's a great question. It's because

the mechanics of inference are entirely different from traditional compute. And that's the core issue GitHub is actually trying to offset here. Okay. How so? So when you execute a traditional database query, the server locates the data, retrieves it, and the transaction is done. Super efficient. Highly efficient. Yeah. But when an AI assistant generates code, it isn't just... looking up an answer in a filing cabinet, it is continuously processing tokens. And tokens are what, like fragments of words? Exactly. Fragments of words or bits of code. It's processing those through billions of parameters. And to do that,

the model has to be loaded into the incredibly fast, highly constrained VRAM of specialized GPUs. Right. So it's monopolizing hardware in a way a standard web request never does. Exactly. For the entire duration that the AI is streaming that code back to your screen, it is locking up prime real estate on a GPU cluster somewhere. Wow. It's performing continuous matrix multiplication. And that requires an amazing... So when GitHub was charging just $29 a month. They were heavily

subsidizing the variable cost of that math. I mean, it was a loss leader to get developers addicted to the workflow. OK, that makes sense. But now that the tool is deeply embedded in enterprise pipelines, you know, they're passing the actual literal cost of hardware utilization directly onto the user. Which creates a massive operational nightmare for anyone running a business. Oh, absolutely. I mean, if I'm a CTO at a startup and AI powered. Suddenly means I have this uncapped metered liability running in the background of every single developer's machine. I can't forecast

my burn rate. You're flying blind. Yeah. Like one month my team is doing light bug fixes and the bill is 50 bucks. But the next month we have a massive crunch to ship a feature. The AI is generating massive blocks of context and suddenly my infrastructure bill is $1 ,000 higher. And that unpredictability is the exact vulnerability that the big tech giants are desperately trying to patch right now. Because they know it's going to scare people off. Exactly. They know metered billing will stall enterprise adoption. In fact,

the Kinsoft source notes that at Microsoft's Build Conference, they spent an enormous amount of time showcasing homegrown, smaller coding models. Oh, interesting. Yeah, they are specifically designing these to be cheaper to run than the big name generalized state of the art models. But wait, are those cheaper models? actually any good? Well, that's the debate. Because, look, if I'm a developer relying on Copilot to catch security flaws or write really complex back -end logic, I don't want some lobotomized, lightweight

AI writing my code just because Microsoft wants to save margins on their GPU clusters. Right. I mean, isn't the industry just... Offering a substandard product to fix a billing problem they created themselves. That is the multibillion dollar tension right now. I mean, they are aggressively pushing small language models or SLS. Right. Under the premise that you don't need a trillion parameter model that knows the history of the Roman Empire just to write a simple Python script. Fair point. So by training highly specialized narrow models, they can reduce the VRAM requirements

and execute the inference much faster and much, much cheaper. But it's risky, right? You are absolutely right to be skeptical because the boundary of where a small model fails and, you know, hallucinates is much narrower. The industry is really trying to thread the needle here. They have to drop the compute cost so they don't bankrupt their users with token fees, but they can't degrade the product so much that users just abandon it completely. Right. And the reason they are so desperate to drop that compute cost brings us to the second major theme of the briefing. Because

the scarcity that's driving these token prices up, it isn't just server availability. No, not at all. It's the physical limits of the planet's infrastructure. Yeah. We have to look at the staggering scale of capital being deployed just to keep this math running. So the Kinsoft briefing highlights this announcement from SoftBank. This is a huge one. It's wild. They have committed up to 75 billion euros to build five gigawatts of AI data center capacity in France. It is almost impossible to overstate how massive five gigawatts

of power consumption actually is. Put that in perspective for us. Well, to ground that in reality, a typical large scale nuclear reactor generates about one gigawatt of power. Wait, really? One gigawatt per reactor. Yeah. Yes. So SoftBank isn't just building a standard server farm. They are proposing an energy footprint equivalent to five nuclear power plants. That is just wow. And it's dedicated entirely to the matrix multiplication required to train and run these models. Here's where it gets really interesting. That explains why they are targeting France, specifically.

Exactly. Because if you need 5 gigawatts of continuous, unbroken baseline power, you can't rely entirely on intermittent renewables. You need the robust nuclear grid that France has built. You need that stability. Yeah, but think about the sheer volume of concrete, steel, cooling systems, and grid -level transformers required to build a 5 gigawatt footprint. It's mind bending. Yeah. I mean, we always talk about the cloud as if it's weightless. Like it's just floating up there. Right. But this is a 75 billion euro industrial megaproject. If we connect this to the bigger

picture, it's the most aggressive mobilization of private capital into physical infrastructure we've seen in decades. Yeah. What we are witnessing is the collision between exponential software demands and linear physical reality. That's a great way to put it. Because you can update software overnight, right? But you cannot build a gigawatt substation overnight. Definitely not. The supply chains for industrial cooling, for high voltage transformers and for the specialized cabling required for these data centers, they're backlogged

by years. So that's why the tokens are so expensive. Exactly. The token prices we discussed earlier are high precisely because this physical infrastructure is fundamentally bottlenecked. And while 75 billion euros is flowing into concrete and copper wire, the Kinsaw briefing contrasts that with a deeply sobering human. This part is really tough. It is. Over at Wikipedia, their volunteer editors actually went on strike. Yeah. And the core issue, they were protesting layoffs among the paid staff,

which they blamed directly on AI driven cost cutting. It is a perfect distillation of the broader capital shift happening across the entire economy right now. Well, on one side of the ledger, you have unprecedented fortunes being poured into energy grids and GPU clusters. And on the exact opposite side, you have the human. element being systematically squeezed out just to free up the capital to buy that compute. I mean, the irony with Wikipedia is incredibly thick here. Yeah. You have a platform that represents the absolute pinnacle of collaborative human curated

knowledge. The best of the Internet, really. Yeah. And the volunteers, people who dedicate their time for free are having to strike to protect the paid administrative roles. It's backwards. They're being displaced by the exact same generative AI models that were literally trained. on the free labor of those very same Wikipedia contributors. Yep. It feels less like an upgrade and more like an extraction, to be honest. That's a highly accurate way to frame the mechanics of what's happening. When an organization automates a human role with an AI tool, the savings rarely just

sit in a bank account. Right. They spend it. They spend it immediately. That capital is almost instantly reallocated into purchasing more software licenses, more compute power, and more token usage. Wow. The Wikipedia strike is a clear indicator. that the infrastructure arms race is not happening in a vacuum. It is being directly subsidized by the restructuring of human labor. I mean, we are literally transferring wealth from payrolls to power grids. Which is a very heavy macroeconomic

reality. But I want to bring this back down to the operational level for you listening right now, because there is an immediate... technical fallout to this massive AI adoption. And it's messy. Super messy. As organizations adopt these expensive tools and desperately try to squeeze productivity out of them to justify that, we wire the AI into our CRMs, our email servers, our cloud drives. We just want all the tools talking to each other. Seamless workflow. Exactly. But the third pillar of the Kinsoft briefing

makes it very clear that this obsession with seamless integration has created a catastrophic security landscape. So we really need to talk about OAuth supply chain attacks. This is undoubtedly the most critical and actionable threat vector mentioned in the sources. The attack surface of the modern enterprise has completely shifted. OK, so anyone managing a SOSDAQ knows OAuth. It's that protocol that allows you to click. integrate with Salesforce or connect to Google Workspace. We all use it. It's incredibly convenient because it allows third -party apps to talk to

your central databases without you having to hand over your actual master password. Right. But the briefing details a major incident with a sales intelligence company called Clue. Right. And Clue is a practically legitimate platform that sales teams use to aggregate data. But the attackers didn't breach Clue because they cared about Clue's internal memos or anything. No, they didn't care about Clue's data. They breached Clue to get access to the Oeth integrations that Clue's customers had authorized. And by hijacking those integrations, the attackers were able to

siphon highly sensitive Salesforce data directly out of the systems of Clue's customers. And this requires us to look really closely at how Oeth scopes are actually implemented in the wild. Okay, bring that down for us. So when a user clicks allow access on one of those pop -ups, they are generating an authorization token. Right. That token dictates exactly what the third -party app is allowed to do. That's its scope. The fatal flaw here is that the industry has a really bad habit of blindly over -provisioning these scopes.

Oh, like giving them too much permission? Way too much. A basic analytics plugin might request full read and write access to an entire CRM database, and a rushed employee will just click approve to get the pop -up out of the way and get back to work. So what does this all mean? To visualize the mechanism here, let's look at how hotel key cards work. Okay. When you check into a hotel, they don't give you the physical master key to the building, right? They program a temporary key card that is supposed to only open your specific room for three days. Right. Very limited scope.

But what we're doing with OAuth is the equivalent of an employee going to the front desk, programming a key card that inexplicably grants permanent access to the penthouse, the security room, and the financial vault. And then handing that key card over to a third -party dog walker or vendor just because the vendor asked nicely. That's exactly it. If that vendor's pocket gets picked, the burglar doesn't have to break your windows or bypass your alarms. They just swipe the perfectly valid key card you created for them. And because it is a valid cryptographically signed token,

it completely bypasses your multi -factor authentication. Wait, really? It bypasses MFA? Yes, it bypasses MFA. It bypasses your biometric logins. The system assumes that because the token is legitimate, the traffic is legitimate. That is terrifying. This is the very definition of a supply chain attack. The attackers target a mid -tier vendor with weaker security specifically to harvest the high -level tokens belonging to their much larger, much more secure enterprise clients.

Which brings me to the detail in the briefing that is just staggering. The victim list. Yes. The victims of this clue breach. The companies that had their core Salesforce data exfiltrated included several dedicated security firms. Yeah, that's the part that hurts. How is that even possible? These are companies selling zero -trust architectures and advanced threat detection. Don't they audit their own integrations? What's fascinating here is it perfectly illustrates the concept of shadow IT and the total collapse of the traditional security perimeter. Shadow

IT meaning stuff happening under the radar. Exactly. In a modern organization, it is rarely the IT department actually provisioning these OAuth tokens. Right. It's a marketing manager connecting a new SEO tool or a sales director integrating a productivity app. Just trying to get their job done. Right. So the security firms were compromised because the technical enforcement of their perimeter was bypassed by human convenience. They trusted a vendor, the vendor got compromised, and the blast radius traveled directly down the supply

chain right into the security firm's database. It renders the entire concept of a firewall almost irrelevant. Pretty much. You were only as secure as the weakest marketing plug -in your intern connected three years ago and forgot about. That's the reality. Yeah. And the Kinsoft briefing emphasizes that this is not an isolated incident. They explicitly tie it to the repeating pattern of the Shiny Hunters threat group, who utilized identical methods to execute massive Salesforce thefts. It's a known playbook now. And to show just how

relentless this broader perimeter collapse is, the source notes two other major breaches from that exact same week. Which ones? Nintendo getting hit by ransomware and losing employee data and Oxford University's careers platform being breached to expose student details. Oh, wow. I mean, whether it's global gaming giants or ancient universities, the data is just bleeding out from every conceivable vector. It's the inevitable result of hyper connectivity. I mean, we. have prioritized seamless API communication above literally everything else. Right. But every

single one of those API connections, every OAuth token is a potential doorway. Just waiting to be opened. Exactly. The attackers have realized that trying to batter down the front door of a highly secure enterprise is a waste of time. It is infinitely more efficient to quietly slip through the side doors that were just left propped open by third party integrations. Which brings us to the synthesis of what this June. June 2026 briefing is really telling us today. Yeah, bringing it all together. We started by looking at the economic reality of the tools we use, right?

The shift to unpredictable metered token billing that forces users to absorb the raw cost of compute. And the sticker shock that comes with it. Then we traced why that compute is so expensive. Because it requires the deployment of massive gigawatt scale physical infrastructure that is actively displacing human labor in the pursuit of capital efficiency. The Wikipedia strike being the prime example. Exactly. And finally, we looked at how our rush to integrate these powerful, expensive systems has basically weaponized our own connectivity

against us through Outh supply chain attacks. The overarching narrative across all these sources, if you look at them together, is friction. Friction, yeah. The era of frictionless tech adoption is over. Every computational choice now carries a distinct, measurable weight, whether that weight is measured in euros, in human jobs, or in security vulnerabilities. Which means we have to move from observation to action. The practical takeaway from listening to this right now is immediate. You need to audit your digital blast radius.

Do it today. Yes, literally today. Log into your core platforms, whether that's Google Workspace, Microsoft 365, GitHub, Salesforce, whatever you use, navigate to the security settings, and pull the list of connected third -party applications. You might be surprised by what you find. You are almost guaranteed to find integrations that were authorized years ago by employees who may not even work there anymore. If an app doesn't have a strict operational necessity today, Revoke its token immediately. Revoke it. You have to aggressively deprecate the lingering access keys

you've left scattered across the supply chain. That kind of aggressive digital hygiene is really the only defense when the perimeter itself is decentralized? But, you know, looking at the entirety of what we've unpacked today, it raises a really profound structural question. Oh, what's that? Well, we are rapidly moving into an environment where every interaction with an interface might incur a metered token charge, right? Where the underlying infrastructure requires city -sized power grids, and where simply connecting two apps exposes your core data to global supply

chain attacks. It's exhausting just thinking about it. Right. So if being plugged in... means absorbing constant financial and security liabilities, will the ultimate luxury in future technology be the ability to completely disconnect and execute your work locally entirely offline? Wow, that is a brilliant angle. If connectivity becomes a liability and cloud compute becomes a metered utility bill, perhaps the most premium feature a piece of software can offer is isolation. Isolation as a luxury. Yeah, because as the Kintop briefing

makes abundantly clear, the plumbing connecting our digital world is expensive, it is leaking, and the bill has finally arrived. It definitely has. Thanks for taking the deep dive with us today. Take a hard look at your integrations, and we'll see you next time.

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. Iris Ohyama - A Suspicious File in the Cloud, 7.7 Million Customers at Risk, and an Online Store That Went Dark

    Nov 19, 202620 minAudio

    Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! Japanese home goods company Iris Ohyama (online store operator Iris Plaza) found a suspicious file in an external...

    Credit the people on this episode

  2. Who Gives A Crap - A Marketing Supplier Breached, Six Days to Tell the Client, and Customer Data That Didn't Need to Be There

    Nov 17, 202619 minAudio

    Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! Australian online retailer Who Gives A Crap told customers on 7 November 2026 that their information was accessed...

    Credit the people on this episode

  3. Last Week in Tech - An Outlook Zero-Day That Leaks Your Password Hash, Chrome's Eighth Exploited Bug, and Nvidia Without Jensen Huang

    Nov 15, 202621 minAudio

    Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! Last week in tech, 7 to 14 November 2026. Microsoft's November Patch Tuesday (10 November) fixes well over a...

    Credit the people on this episode

  4. Chainalysis - From "Non-Sensitive" to "All Users": How a Breach Notice Changed in Five Days

    Nov 12, 202619 minAudio

    Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! Chainalysis disclosed on 30 October 2026 that an external actor using "offensive agentic techniques" accessed...

    Credit the people on this episode

  5. NSW Health - A Dental Voucher Website, Four Gigabytes Taken, and the Providers Whose Details Were Exposed

    Nov 10, 202623 minAudio

    Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! NSW Health reported unauthorised access to its Oral Health Fee For Service Scheme website on 29 October 2026;...

    Credit the people on this episode