Last Week in Tech – SpaceX Buys Cursor for $60B, Washington Pulls Anthropic's Top Models, and the Klue Breach Spreads
20 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: SpaceX makes the largest startup acquisition in history, buying AI-coding company Cursor for around $60 billion in stock - days after its own IPO. Washington uses export-control powers to force Anthropic to pull its most powerful models on national-security grounds. The Klue OAuth breach we flagged a fortnight ago balloons to nearly 200 companies, including a who's-who of security vendors. And world leaders sit down with AI's biggest names at the G7, as SoftBank pledges €45 billion for French data centres. Trying to work out what all this AI consolidation means for your own tech stack? Visit www.kinsoft.com.au to talk through your security and IT needs. Sources: CNBC; TechCrunch; Fortune; SecurityWeek; BleepingComputer; Global Banking & Finance.
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.
Imagine spending, I mean, literally millions of dollars integrating the world's most advanced AI into your business. Right. Your development team loves it. Your daily operations rely on it. Your revenue is practically tied to it at this point. Yeah, it's a load -bearing pillar for the company. Exactly. And then overnight, you wake up to find a government has legally forced the provider to just pull the plug. Just shut it all down. Yeah. The software hasn't crashed. The servers aren't down. It has simply been seized
and your business is left completely in the dark. Which is a terrifying thought. It really is. So welcome to our deep dive. Whether you are catching up for a major tech strategy meeting this week or you are just, you know, insanely curious about the fragile digital landscape we are all navigating, we built this deep dive specifically for you. We really did. Today, we are pulling apart a single incredibly dense update from Tech Talks with Kinsoft. This is... dated late June 2026. And our mission here is to just cut through the daily noise and examine a massive rapid shift
happening right now regarding who actually controls the digital tools your business relies on every single day. So, OK, let's unpack this. Yeah, it really is a fascinating collision to unpack. We are looking at a critical moment in tech history where incredibly rapid unchecked innovation is suddenly smashing head on into, well, extreme corporate consolidation on one side and then forceful government intervention on the other. Right. It is just a fragile new reality for anyone who builds, manages or simply uses technology today. And I mean, nothing screams extreme corporate
consolidation quite like the sheer scale of the numbers in this first story. Oh, absolutely. Let's look at what happens when tech giants decide to swallow the very tools that build the tech industry. So on June 16th, 2026, just days after pulling off its own blockbuster IPO, SpaceX filed to acquire any sphere. The company behind Cursor. Exactly. For context, for you listening, AnySphere is the company behind that massively popular AI coding tool called Cursor. And the price tag for this acquisition is an all -stock transaction
worth around $60 billion. $60 billion. I mean, $60 billion is a staggering figure. It's wild. It is, especially when you consider that, according to the source material, this makes it the largest acquisition of a venture -backed startup in history. Wow. In history? Yeah. But when you look at the underlying mechanics of any sphere's business, you start to see the strategic math behind it. Cursor was carrying somewhere between $2 .5 and $4 billion in annualized revenue. Okay, so a
massive cash engine already. Exactly. Yeah. And more importantly, they had over a million paying users. And it's not just Cursor operating at this scale, right? The source points out that nearly every serious AI coding tool is now sitting under a mega corporate umbrella. Yeah, the whole board has been bought up. Right. You've got Copilot owned by Microsoft. You've got Windsurf and Codex orbiting OpenAI. You have Claude Code over at Anthropic. And now Cursor is absorbed by SpaceX. It's a clean sweep. It really feels like the era of the railroad barons where the independent
companies laying the tracks are suddenly being bought up by three or four massive empires. That's a really good way to put it. But, I mean, push back here for a second. Companies grow, right? And startups get acquired by bigger fish all the time in Silicon Valley. That's the whole model. Sure. So is this actually bad for the average developer or is it just the natural evolution of tech? Well, what's fascinating here is that. It entirely shifts the power dynamic in a way that introduces massive vendor risk for the end user. How so? Let's explore how a software development
team actually functions today. When a business decides to use an AI coding tool like Cursor or Copilot, they don't just casually try it out on a weekend. Right. It's not like downloading a new notes app. Exactly. They integrate it into their core systems. They standardize their coding workflows on it. Their developers build intricate habits and processes around how that specific tool functions, you know, how it suggests code, how it debugs. Okay, yeah, I see that. So if my entire engineering team spends a year learning the exact quirks and shortcuts of one specific
AI tool, changing it out isn't just swapping a piece of software. No, not at all. It's practically retraining the whole department. It becomes pure muscle memory. Yes. And that muscle memory is what creates absolute lock -in. And this is where we have to understand the mechanics of an API and application programming interface. Right. In the past, you bought software in a box, right? Yeah, you install it and it's yours. Exactly. Right. But today, you are just running access to a data pipe hosted in the cloud. Once that
tool is deeply embedded in your workflow, your leverage as a buyer drops to near zero. Because you can't just leave. Right. If the independent startup you initially signed a flexible contract with is suddenly absorbed by a $60 billion titan like SpaceX or Microsoft, the rules of the game change. You have to know who ultimately owns the tools you are betting your business on. Because your leverage just shifted dramatically. Entirely. If that new mega owner decides to double the price or drastically alter the privacy policy
regarding how your proprietary code is fed back into their machines to train future models, I mean, you have very little recourse. The market is shrinking to a handful of giants and they hold all the leverage. Wow. So you might think you are working with a plucky, independent startup, but you could wake up tomorrow and find out you are actually just a tenant on one of the mega empire's land. And you are subject to the rent hikes. But as powerful as these corporate empires are, there is a higher power that can step in and remind them they aren't completely in charge.
Yes, there absolutely is. So you might be renting assets from one of these mega empires. But what happens when a government decides to seize the building? Yeah, this is where the landscape abruptly shifts from corporate strategy to national security. Right. And it introduces a level of operational risk most businesses are completely unprepared for. So here's where it gets really interesting. In early June, after Anthropic launched its most powerful new models, the U .S. Commerce Department stepped in. And they did not mess around. No, they didn't. They issued an export control directive
ordering Anthropic to suspend worldwide access to those top tier models. Worldwide. Worldwide. And not just for the general public either. This suspension even applied to Anthropic's own foreign national staff working inside the company. Which is just a massive move. The mechanism behind this unprecedented shutdown, according to the source material, was a discovered jailbreak. OK, a jailbreak. Right. The government realized these models contained a vulnerability that could potentially turn them into unrestricted cyber tools. Wow. Yeah. To understand why the Commerce
Department reacted so aggressively, we need to quickly define what we mean by frontier models. Right, because a frontier model isn't just a slightly smarter chatbot. Exactly. It is a highly advanced AI system that matches or exceeds the capabilities of the most powerful existing models on the market. Because they are pushing the absolute boundaries of what AI can do, they introduce entirely novel security risks. And one of those risks is that jailbreak mechanism. Large language models are fundamentally designed to predict
text based on the prompts they receive, right? Yeah. They have safety guardrails built in to prevent them from, say, writing malicious code or generating instructions for a biological weapon. Right. The stuff you definitely don't want them doing. Exactly. Yeah. But a jailbreak, often achieved through a technique called prompt injection, uses adversarial, highly specific language to confuse the model. It essentially tricks it into ignoring its safety training. Oh, I see. Yeah. And once those rails are bypassed, an advanced
frontier model can theoretically be used to write zero -day exploits or automate cyber attacks at scale. So the Commerce Department viewed this text generator not as a software product, but as a potential cyber weapon. Just to be clear on the facts of the dispute here, Anthropic did comply with the directive. They did, yes. They took the models offline globally, but they also publicly disagreed with the government's action, pointing out that this exact same weakness affects other frontier models out there, not just theirs. Right. They felt singled out. Yeah. And, you
know, we are looking purely at the facts of the event, not taking sides here. But if I'm a business leader. who just built my company's operations on one of these load -bearing ai models and it vanishes overnight i mean what do i even do is this just the new digital act of god well if we connect this to the bigger picture it absolutely is a digital act of god yeah and it forces a complete rewrite of how businesses handle disaster planning and continuity risk. Oh, so. Think about
the physical mechanics of traditional software. If you bought an on -premise database system 10 years ago, you installed it on physical servers sitting in a room you controlled. Right, under lock and key in your own office? Yes. If the government sanctioned the creator of that software or the company went bankrupt, your software didn't suddenly uninstall itself from your hard drives. The lights stayed on. Right. I might not get the next security patch, but my customer service team could still log in the next morning. Exactly. But today, with frontier AI models, you don't possess the software. You are renting a continuous
stream of data from a cloud server. Just a pipe. Just a pipe. A year ago, the idea that a government could step in and legally force a provider to sever that data stream, instantly breaking the core functionality of businesses all over the world. That simply wasn't a threat model most IT departments were seriously considering. No, you'd think about server outages, not federal seizures. Right. But this event proves that a frontier AI model can be switched off globally overnight on national security grounds. So if
your operations rely on AI to function, if that AI is load -bearing for your company, you suddenly have to ask yourself, what happens if your primary model becomes a geopolitical football tomorrow morning? You need a backup plan. Yeah, a backup plan that doesn't just rely on a different tool, but perhaps doesn't even rely on the same national jurisdiction. It is a stark reminder that while megacorporations control the tools, sovereign power still dictates the ultimate terms of service. It highlights a massive vulnerability at the very top of the tech food chain. But, you know,
the irony here is staggering. Oh, for sure. We just talked about massive $60 billion corporate buyouts and top tier models that the government can turn off overnight like cyber weapons. But you don't even need a federal directive or a state sponsored quantum hack to bring a company to its knees today. Not at all. According to the Kinsoft update, the most widespread threat right now isn't futuristic at all. It's basic plumbing. You just need a sloppy digital handshake. The deeply unglamorous reality of tech hygiene. Exactly. So let's look at the Clue breach. Clue
is a market intelligence software vendor, and they were compromised not by a sophisticated zero -day exploit, but through a forgotten integration credential. A forgotten credential. Yeah. And the fallout is massive, with the breach spreading to nearly 200 organizations. And to understand the sheer scale of this, we really need to look at how this mechanically happened. Okay, let's hear it. The attackers managed to harvest what are known as OAuth tokens. Now, for our listeners, OAuth is the industry standard protocol that allows one application to access data in another
application without you ever having to hand over your actual password. Okay. So when you click a button that says sign in with Google, or when you authorize a third -party app to connect to your Salesforce environment, you are using O. Oh, right. So what does this all mean? It's like a hotel valet key. A valet key, exactly. You hand the parking attendant a key that allows them to turn on the engine and park the car, but it physically prevents them from opening the trunk or the glove box where you keep your valuables. O is a digital valet key. That is
a perfect analogy. And so Clue was handed a digital valet key to connect seamlessly into its customers' Salesforce environments to pull the data it needed for market intelligence. Right. And that valet key analogy explains exactly how the attackers bypassed all the expensive front door security. Because they just went for the keys. Yes. When an oath connection is made, an access token is generated and stored in a database on a server. Some of these tokens are incredibly long -lived. I mean, they don't automatically expire unless
someone manually revokes them. Wow. They just stay active forever. Essentially, yes. The attackers didn't try to pick the complex locks on the front doors of these 200 companies. They broke into the valet stand. Stole the master ring of digital valet keys and quietly used those existing trusted tokens to query the customer's CRM data directly. That is wild. They bypassed the firewalls, the biometric logins, the complex password requirements, all because they had a pre -approved digital pass. Yeah. And this wasn't just hitting small,
unsuspecting companies without IT departments, right? No, not at all. The Kinsoft to SourceNuts, the victim list is a who's who of elite security firms. We are talking about hundreds. Interest tenium, recorded future. These are organizations whose entire business model is catching cyber threats, and they were compromised through a vendor's forgotten token. It's like spending a million dollars on a state -of -the -art bank vault, but leaving the spare key under the doormat for the window cleaner. Exactly. And this raises an important question, because it fundamentally challenges how we view digital security. We spend
millions of dollars and countless hours worrying about the strength of the vault door. But this incident proves the real danger lies in the trusted side channels. Huntress, Tanium, Recorded Future. These organizations are filled with world -class experts. If they can get caught by a compromised OOOTH token, anyone can. Because it isn't a failure of intelligence or a lack of sophisticated security software? No, it is a pure failure of hygiene. The takeaway here is the compounding nature of
vulnerability. One single... trusted app connection can become a doorway into hundreds of organizations. Wow. You can build the most secure internal network in the world, but the moment you grant a third -party app an Outh token to reach your data, your security posture is instantly lowered to match that third -party vendor's security. You're only as strong as your weakest integration. Exactly. If they forget to rotate a credential or if a token sits on a server untouched for two years, you are entirely exposed. So the actionable advice here for you listening right now is to go audit
your systems today. Look at which third -party applications have OAuth access to your Salesforce, your Microsoft 365, your Google Workspace. Yes, please do that. If you see an integration that is stale or an app your team hasn't actively used since 2024, revoke the permission immediately. Cut the connection. Take the valet key back. It is tedious administrative work, I know. But as the clue breach clearly demonstrates, it is arguably the most critical defense mechanism you have. It really paints a picture of an incredibly
fragile landscape. I mean, megacorporations are buying up the foundational tools. Governments are shutting down the frontier models overnight. And forgotten tokens are exposing the world's top security firms. It's a lot to take in. It is. So taking a step back, how is the rest of the world reacting to this on a macro level? Well, the contrast on the global stage is what really stands out on the Kinsoft update. At the G7 summit in France, world leaders held an unprecedented working session. They actually brought in the top executives of OpenAI, Anthropic, and Google
DeepMind. Okay, so the absolute biggest players. Yes. Getting all of those major AI figures in the same room with the G7 leaders is a... highly unusual and major diplomatic event. Right. You'd expect some massive ironclad regulations to come out of that. But the source says they landed on a package of largely voluntary AI commitments focused on child safety online. Yes. Mostly voluntary. Wait a minute. We just spent 10 minutes talking about. The U .S. Commerce Department forcefully
shutting down anthropics models globally because they treat them like literal cyber weapons. Right. And now the G7, the most powerful international table in the world, is rolling out voluntary non -binding standards for child safety. That feels like a massive disconnect in how governments are treating this. Are they treating this as a hard, immediate threat or just making soft suggestions? You've zeroed in on the exact contradiction that defines this current era of tech regulation. We are watching a split reality play out. out in real time. On one side, when it comes to international
diplomacy and trying to write a global rulebook, you have these soft emerging standards. Governments at the G7 level are moving slowly, relying on voluntary commitments because achieving international consensus on hard law for a technology evolving this rapidly is practically impossible. Right. Diplomacy is a slow machine. Extremely slow. So they settle for handshakes and diplomatic press releases while they try to figure it out. Meanwhile, the tech itself isn't waiting around for them. Exactly. On the other side of the equation,
the physical reality of this technology is moving at light speed, and it is anything but soft. To run these frontier models, you need immense computing power. You need hardware. Heavy hardware. The source mentions that right around the time the G7 was agreeing to voluntary standards, SoftBank pledged around 45 billion euros to build AI data centers in France. 45 billion euros just for data centers? Yes. This is what we call the capex arms race capital expenditure. While the global regulatory rules remain soft and voluntary, the
financial and structural reality of the AI industry is cementing itself into the ground. 45 billion euros isn't buying software. It is buying massive tracts of land, securing water rights for cooling systems, laying thousands of miles of high -voltage power grids, and installing specialized silicon chips. It's incredibly physical. Yes. The governments might be moving slowly with regulations, but the infrastructure required to run these virtual tools is becoming a permanent, massive footprint on the physical world. It really brings into
focus how complicated your world is getting if you are a business leader trying to build on top of these systems. It's a minefield. It is. To summarize the massive arc we've traveled today for you, the listener, the actual tools your developers are using to build your company's future are rapidly shrinking into the hands of a few mega giants like SpaceX and Microsoft, and that is severely limiting your leverage. Yes. At the exact same time, governments are realizing they have the power to instantly pull the plug on those load -bearing AI models. under the banner of national security. The kill switch.
The kill switch, right. And yet... But despite all this high level corporate consolidation and sovereign power, your entire massive expensive operation can still be completely brought down by a single forgotten digital valet key left open by a third party vendor. It just demands a level of continuous vigilance and strategic planning from business leaders that simply wasn't required a decade ago when software lived on a disk on your desk. No, not at all. So our call to action for you enthusiastically reminds you
to stay patched. Stay deeply skeptical of what you're plugging into your network and go check those OAuth tokens. Digital hygiene is just as important as the strength of your firewalls. Absolutely. And to leave you with one final thing to mull over. Building on all the source material. He's due. We've seen how the entire digital infrastructure, from the code your developers write to the 45 billion euro data centers powering it, is consolidating into just three or four mega corporations. Yeah. If these entities now hold the keys to tools that can single -handedly trigger global national
security emergencies, are we moving toward a future where these tech giants essentially have to act and be treated as nation states themselves? Man. When the company's laying the digital tracks have the capital and the power of a country, you really have to wonder what kind of world we are all renting space in. Thanks for joining us on this deep dive. Keep questioning the systems you rely on, and we will catch you next time.
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...
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...
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...
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...
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;...