Last Week in Tech - The Biggest Patch Tuesday Ever, a 10.0 With No Patch to Apply, and Oracle's $664 Billion Round Trip
20 minAudio
Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! Last Week in Tech for Monday 14 September 2026, covering 7 to 13 September. Microsoft's largest Patch Tuesday ever (8 Sep): around 970 CVEs - 974 per Microsoft's own release note, 964 per Tenable, 966 per BleepingComputer. Two were already being exploited, both Windows privilege escalation, both CVSS v3.1 7.8: CVE-2026-85880 (ALPC heap buffer overflow) and CVE-2026-81963 (Windows Update Stack link-following), per Tenable the first Update Stack flaw ever caught being exploited. Both are local escalation, not initial access. ZDI's Dustin Childs rates Exchange CVE-2026-55007 (8.1, malicious Visio attachment) the month's most important patch, and calls Windows DNS Server CVE-2026-69730 (9.8) the spiritual successor to SigRed. September's Windows Server updates break Remote Desktop Services - test first. Magento and Adobe Commerce "StyleSmuggler": CVE-2026-75650, CVSS v3.1 10.0, unauthenticated RCE. Exploited from 4 Sep, disclosed by Sansec 5 Sep, patched 7 Sep (APSB26-146), added to CISA KEV 8 Sep. A Magento-written log file is poisoned with PHP, then executed when Magento renders the standard Payment Transaction Failed Reminder email - nobody needs to open it. Sansec's first victim ran 2.4.6-p15 with July and August updates applied, the current patch level for that line; Disrex Group confirmed two more, one on 2.4.8 running Sansec's own protection product. Patching does not remove an installed backdoor - any store online between 4 Sep and patching needs a compromise assessment, a session flush, and rotation of the encryption key, admin passwords and all env.php credentials. Ten CVEs into CISA KEV in three days, almost all edge devices. 9 Sep: Cisco Secure Firewall Management Center auth bypass to root, CVE-2026-20079 (10.0), with Cisco confirming exploitation from August; Citrix NetScaler auth bypass CVE-2026-19490 (9.3); Fortinet heap overflow CVE-2025-25249 (7.3), linked by SOCRadar to a PivotC2 campaign that infected 178 devices. 10 Sep: MikroTik RouterOS CVE-2026-67277 and CVE-2026-86060 - CERT Polska warned of full administrative takeover via internet-exposed SSH without authentication, calling it MikroTrick. Zero-day status unverified. Oracle Q1 FY2027 (10 Sep): revenue US$19.3bn (+30%), cloud US$11.6bn (+62%), infrastructure US$7.4bn (+121%), backlog US$664bn against roughly US$640bn expected, with 300,000+ GPUs delivered. On 11 Sep shares rose as much as 7.8% intraday, then reversed to close around 2% down. S&P estimates roughly half the backlog is tied to OpenAI, calculated against the prior US$638bn figure; Moody's has flagged counterparty risk. Apple (9 Sep): the iPhone Duo foldable starts at US$1,999, on sale 23 October. iPhone 18 Pro from US$1,199 and Pro Max from US$1,299 ship in 65+ countries from 18 September. All run the A20 Pro, Apple's first 2nm chip. There is no standard iPhone 18 this cycle - the affordable models move to spring 2027, pushing mid-tier fleet refreshes out two quarters. The AI cost gap: DeepSeek's V4.1-Flash (10 Sep) has MIT-licensed open weights at US$0.15 per million input tokens and US$0.60 per million output off-peak, though its benchmarks are vendor-published and not independently reproduced. The Information reported on 6 September that Anthropic has signed compute agreements worth up to US$517bn - an estimated maximum across many deals, not published by Anthropic. Coming up: Wednesday, a Sydney edtech breach affecting over a million people. Friday, the developer-tools vendor that missed its own patch. Visit www.kinsoft.com.au to talk through your security and IT needs. Sources: Microsoft Security Update Guide; Tenable; BleepingComputer; Sansec; Adobe APSB26-146; Disrex Group; CISA KEV catalogue; CERT Polska; Oracle investor relations; The Information.
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 a burglar walks up to your house, right? They don't pick the lock. They don't throw a brick through a window. They just wait. Exactly. They just wait for your smart doorbell to automatically send you a notification. And the mere act of your doorbell generating that alert causes your front door to just fly wide open. You never even pressed a button. It's a really visceral image. But it honestly perfectly captures the reality of the digital landscape right now. The very defenses we invest in. are becoming the doorways.
And we're going to get into the specific mechanics of exactly how that is happening in today's deep dive. We're unpacking a single just densely packed source today. It's a mid -September 2026 industry update from Kinsoft called Last Week in Tech. Right. And our mission here is to extract the critical, you know, the need to knows for anyone who relies on digital infrastructure, which, let's be honest, that is every single one of us. Yeah. We are looking at a tech world that is currently buckling under extreme stress. I mean, from attackers finding mind -bendingly
clever ways to slip past our perimeters to the sheer crushing weight of the money being poured into artificial intelligence right now. And if you are listening to this on your commute, maybe you're an IT director or maybe you run an e -commerce brand or you're just trying to understand where the Internet is headed. These hidden mechanics, they dictate your daily friction points. Absolutely. These aren't abstract concepts. These vulnerabilities, these financial shifts, they're the reason you might have to, I don't know, authorize weekend overtime for your team or even rewrite your entire hardware budget mid -quarter. Okay, let's unpack
this. Starting with that doorbell analogy and how it applies to online retail. an online store on Magento or Adobe Commerce. The source material essentially issues a red alert. A massive red alert, yeah. In early September, attackers started exploiting a zero -day vulnerability dubbed Style Smuggler. And I want to make sure I understand the sheer severity of this because it has a CVSS score of 10 .0. That's the common vulnerability scoring system. And 10 is the absolute maximum, meaning it requires zero authentication and an
attacker can execute code remotely. But what I don't get is how does it trigger without anyone actually clicking a malicious link? The mechanism is genuinely brilliant in a highly malicious way. Typically, an attacker tries to force malicious code through an input field, like a login box. Like typing code into a password field. Exactly. But here. They completely bypass that. They poison a standard log file that Magento naturally writes to just to keep track of its own activity. And they plant PHP code inside of it. Then they just
trigger a completely mundane automated function. Specifically, it's a payment transaction failed reminder email. Wait, so let me make sure I have this straight. The email doesn't even need to be opened by an employee? Nope. Just the act of the systems thinking about rendering that normal message triggers the hack. Yes, exactly. When Magento simply attempts to assemble that email, it pulls from that poisoned log file and the malicious code executes. It's zero click in the purest sense. Nobody clicks a link. Nobody opens an attachment. Wow. And what drops into your system after that automated door opens is
a highly sophisticated 1 .9 megabyte payload. And it's written in Rust. Which is super fast, right? Yeah. Rust is a programming language known for being incredibly fast and notoriously hard to reverse engineer. This program disguises itself as a normal Linux background task. It hides in a hidden directory above what's called the web root. Meaning what exactly? Basically, it stores itself outside the folder where your actual website files live. So your typical security scans. They completely miss it. Oh, wow. And then it sets
up a scheduled job to resurrect itself every five minutes. Yeah. The source notes one store where this resurrection job was added over 1 ,700 times. An admin would delete it and it would just regenerate in less than a second. It's relentless. Yeah. But the most alarming detail for anyone managing these stores isn't just the profistence. It's the fact that patching the software was entirely irrelevant to stopping the current infections. See, I saw that in the notes and assumed it was a typo. How can a fully updated store still be compromised? Because the vulnerability allowed attackers to plant their persistence mechanisms
deep in the operating system before the patch was even available. Oh, I see. Yeah, stores running version 2 .4 .6, patch 15, which is the most current patch level they were hit, even a store on the newer version 2 .4 .8 running a paid commercial protection product got compromised. So patching does nothing if they're already in. Exactly. If your store was online and exposed between September 4th and September 7th, you need a full compromise assessment, not just a software update. You literally have to assume they are already inside. So you have to burn it all down, basically.
Pretty much. You have to flush user sessions, rotate encryption keys, change admin passwords, everything, and be warned about the detection trap here. What's the trap? If your security team just searches for the exact text string that the initial security reports published, they will fail to find the malware. Because the attackers changed the fingerprint. Exactly. The infections quickly moved to a different system log directory. And that static marker string changed within 24 hours. You have to hunt for the behavior of the malware, like a hidden process
constantly restarting, not just scan for a static line of code. And this Magento flaw is just, it's one piece of a much larger crisis happening at the edge of corporate networks right now. Oh, yeah. It's everywhere. The source points out that the American Catalog of Exploited Vulnerabilities added 10 different edge device entries in just three days. We're talking about a massive flaw in Cisco Secure Firewall Management Center giving attackers root access. Yep. Flaws in Citrix Netscaler, Fortinet, and this wild microtrick vulnerability
taking over microtrick routers via SSH without any authentication at all. I mean, why are the devices meant to protect us suddenly the most vulnerable things on the network? It is the cruelest irony in modern cybersecurity, really. For years, the fundamental rule of defense has been to reduce your attack surface. You know, shut down open ports, turn off unnecessary services. Right. But with a firewall manager or a VPN gateway or a load balancer, you cannot reduce their attack surface because the attack surface is the product.
Right. They have to face outward. They are designed to stand at the absolute edge of your network and face the public internet 24 -7. And because these devices are getting exploited before the vulnerabilities are even publicly disclosed, the fact that you applied a patch today tells you absolutely nothing about whether someone already walked through that door last week. So the strategy has to change. The new mandatory rule for IT strategy is patch, then hunt. But fixing the software that faces the outside world only solves half the problem. If attackers slip
past that perimeter firewall, what are they walking into? Because this week, they are walking into a Windows environment that requires nearly a thousand separate bandages. It's really bad. Let's look at Microsoft's historic September 2026 patch Tuesday. The volume of updates here is staggering. It is a massive payload. I mean, the running total for Windows vulnerabilities in 2026 is now roughly 2 ,760. Which is insane. Let's put that into perspective. The previous record year for vulnerabilities was 2020, which
had 12 ,245. We have more than doubled the worst year on record, and we still have a quarter of the year left. Mixed into those hundreds of patches are two actively exploited zero days. They're both Windows privilege escalation flaws carrying a CVSS score of 7 .8. Right. I have to push back here, or at least ask for some clarity on how security teams prioritize these things, because we just spent 10 minutes treating a 10 .0 CVSS score from Magento like a five alarm fire. Why is a 7 .8 being treated with the same extreme urgency by these security researchers? Like,
what does privilege escalation actually mean in plain English for a business owner? What's fascinating here is that the raw severity score can sometimes mask the real world operational danger. A 10 .0 is the flaw that gets the attacker into the building. But a 7 .8 privilege escalation flaw is what they use once they are already inside. Think of it like a corporate badge system. A phishing email might get an attacker a visitor badge, giving them access to the lobby. Right. They're in, but they can't do much. Exactly.
But a privilege escalation flaw is a glitch in your badging software that lets that visitor walk up to a kiosk and instantly reprint their ID to say CEO. Suddenly they have access to the server room, the executive suites, everything. Okay, that makes total sense. It's the mechanism that takes a minor contained breach, like, say, a single compromised laptop in HR, and turns it into a company -wide ransomware catastrophe. Precisely. It gives them system control. And while IT teams are triaging that massive list of patches, the source highlights two others
that demand immediate attention. What are those? First, an exchange server flaw with an 8 .1 score. This triggers remote code execution simply by a user receiving a malicious Visio diagram attachment in an email. Wow, just receiving it. Yeah. Second, a 9 .8 score flaw in the Windows DNS server, which researchers are calling the spiritual successor to the infamous cigarette vulnerability. DNS is basically the phone book of the Internet, right? It translates website names into IP addresses. Yes. And if you run an Internet -facing Windows
DNS server, a flaw like this acts like a corrupted switchboard operator. An attacker can hijack that server and misdirect all your internal and external network traffic wherever they want. So you think you're going to your bank, but you're actually going to their fake site. Exactly. That is a drop everything and fix it moment. But there is a massive warning from the source material regarding this specific patch Tuesday. You absolutely must test these updates in a sandbox environment before deploying them to your live servers. Yeah, I read this part and genuinely felt bad for system
administrators. The September Windows server update actually causes a deadlock between the remote desktop protocol and the local session manager. I know that sounds super dense, but what is the actual mechanism happening there? It's essentially a software traffic jam that can't be resolved. Think of it like two security guards at a single narrow door who both refuse to move until the other one steps aside. So they just stand there forever. Exactly. Nothing can pass. When this deadlock happens, it completely
breaks remote desktop services across your entire environment. Meaning, if an IT director pushes this update blindly just to secure the network, they accidentally lock their own IT team out of the very servers they are trying to manage remotely. Exactly. You secure the building by supergluing all the locks shut with your team on the outside. That's terrifying. You really cannot afford to skip the testing phase, even when the pressure to patch a thousand vulnerabilities is this high. Right. Well, if your software is this brittle, you'd hope your hardware provides a solid foundation, or at least that your company
can afford to keep it updated so you can run all these new security protocols. But the hardware reality check that just hit corporate IT departments following Apple's latest event completely rewrites that budget math. It was a massive hardware drop, sure. But the real story for enterprises is hidden in what didn't happen. The event was undeniably flashy. I mean, we finally got the foldable iPhone Duo opening up to a 7 .8 -inch screen. Pricing starting at $1 ,999 and scaling all the way up to $3 ,199. Which is a lot for a phone. A whole
lot. Plus, they rolled out the iPhone 18 Pro and Pro Max featuring that new 2 -nanometer A20 Pro chip alongside a whole suite of new watches and AirPods. It's just an incredible amount of new tech. If we connect this to the bigger picture, though. Those ultra -premium specs are creating a logistical nightmare for corporate device fleet managers. How so? The critical omission in that lineup is that there is no standard iPhone 18 in this release cycle. Apple has delayed the more affordable baseline models to spring 2027. So how does a six -month delay on a standard phone upend a corporate IT budget? Well, imagine
you're an IT director for a mid -sized company, maybe 1 ,000 employees. You budgeted $800 per device for a mid -tier hardware refresh this quarter. Your employees' current phones are aging out of their security support windows, which, as we just discussed with all these flaws, is incredibly dangerous right now. You definitely want supported devices. Exactly. But because the standard model is delayed, your assumptions just got destroyed. Your options right now start exclusively at the pro -pricing tiers, which
might be $1 ,200. You are suddenly looking at a $400 ,000 deficit in a quarterly budget you finalized months ago. Wow. So you either have to go to the CFO and beg for half a million dollars of unplanned capital expenditure, or you force your employees to run vulnerable aging hardware for another six to eight months and just pray you don't fail a compliance audit. That is the exact corner thousands of businesses have just been painted into. It's a huge problem. I mean, a $400 ,000 hardware deficit is enough to get an IT director fired. But when we zoom out to
the macroeconomic scale of the tech industry right now, that kind of money is literally a rounding error compared to what is being spent to build the infrastructure for artificial intelligence. Oh, entirely. It's not even a drop in the bucket. I was looking at Oracle's Q1 FY2027 financials in the source material, and the stock market's reaction made zero sense to me. They posted $19 .3 billion in total revenue. Their infrastructure cloud revenue was up 121%. When the numbers dropped, the stock surged almost 8 % during the day. But
then it completely plummeted and closed 2 % down. What spooked the market so badly? It was a single line item that revealed the staggering, almost unimaginable scale. scale of the AI infrastructure boom, and more importantly, the terrifying risk holding it up. OK, what was the line item? The market initially cheered because Oracle revealed a backlog, the remaining performance obligations of $664 billion. $664 billion. Yeah. With a B. Yes. Heavily driven by $30 billion in new AI cloud contracts signed in just one quarter. alongside
the delivery of 300 ,000 GPUs. But the market reversed course violently because they realized the extreme counterparty concentration risk. Meaning who they're doing business with. Right. Roughly half of that massive $664 billion backlog is tied to a single entity, OpenAI. Wait, so Oracle is basically betting the entire farm on one horse. It raises a vital question about whose balance sheet is actually underwriting these massive physical data centers. Look at the mismatch in scale here. OpenAI's estimated annualized
revenue was around $12 billion. Okay. Yet they are making infrastructure commitments in the hundreds of billions of dollars. Meanwhile, to fund building these physical data centers to house all those GPUs for OpenAI, Oracle sold $20 billion in stock. They are currently running a negative free cash flow of $5 .4 billion, and they are guiding a net cash outlay of roughly $70 billion for the year. So Oracle is spending tens of billions in actual present -day cash to pour concrete and buy servers based on a half -trillion -dollar IOU from a startup that only
generates $12 billion a year in revenue. You've just identified exactly why the stock dropped. If OpenAI's subscriber growth slows down, or if the next generation of their models doesn't deliver the expected commercial leap, Oracle is left holding the bag on tens of billions of dollars of highly specialized depreciating Physical hardware. That is, I mean, that's insane. It is the ultimate lesson in supplier concentration risk. And this top -end infrastructure spending looks even more precarious when you contrast it with the reality of what's happening at the bottom end of the AI market. On one hand, you
have Anthropic reportedly signing compute agreements worth up to $517 billion over an 11 -month period. consuming 14 .8 gigawatts of power. Which is astronomical. Right. To put that in perspective, 14 gigawatts could power a small country. But on the exact same week, a Chinese AI company named DeepSeek releases their V4 .1 flash model. And the economics of that DeepSeek release fundamentally challenged the multi -billion dollar thesis. So the source notes it has open weights and an MIT license. Let's translate that for a second. Open weights means the actual brain of the AI,
the underlying neural network, can be downloaded by anyone. And the MIT license means... a business can use it commercially with almost no restrictions. Right. It's totally open. And it scores near the absolute top frontier level on coding benchmarks. But the cost to run it, 60 cents per million output tokens, that's 60 cents to generate roughly 750 ,000 words. The gap between the top end and the bottom end of the AI market is, it's no longer a gap. It's a chasm. Yeah. The architectural efficiency is happening in the open source community,
utilizing techniques like routing queries to smaller, specialized sub -models. Instead of activating a massive trillion parameter brain for every simple question, they're driving cost to zero. Now, there are major caveats, of course. Like, well, if you use a model like DeepSeek. You have to navigate complex data residency laws. If you don't host it yourself, your corporate data is bouncing through an overseas API. And you actually need the internal engineering talent to download those open weights and spin up the servers in -house. True, but if a company does
have that talent, the build versus buy math for enterprise AI has totally inverted. I mean, a business processing millions of legal documents or analyzing massive data sets can suddenly do it for absolute pennies using highly capable open models rather than paying the astronomical frontier infrastructure prices demanded by the major labs. Which puts immense existential pressure on those heavily leveraged half trillion dollar infrastructure bets we just talked about with Oracle, OpenAI and Anthropic. The competitive landscape and the fundamental value of a generated
word is shifting beneath their feet while they are still busy pouring concrete for their new data centers. So what does this all mean? We've covered a massive amount of ground today, and it really paints a picture of a tech landscape experiencing absolute whiplash. In just a few days, we are dealing with zero -click Magento payloads that completely bypass all software patches. We are staring down historically massive Microsoft updates that act like a digital traffic jam, literally locking IT teams out of their own servers. We have Apple's supply chain logistics
quietly forcing corporate fleets into budget -busting hardware. upgrades. And we are watching hundreds of billions of dollars being leveraged on AI infrastructure that relies on the continued, flawless success of a single customer. The throughline connecting all of this is that blind trust is the single biggest vulnerability a business can have today. Whether that is blindly trusting that a software patch perfectly protects your network perimeter without you actually hunting for the intrusion, or blindly trusting an automated operating system update won't break your internal
architecture. or blindly trusting that massive AI suppliers have the financial gravity to underwrite the future of your company's technology stack. The digital perimeter isn't just crumbling. The foundation it sits on is actively shifting. You have to verify everything. You have to assume breach as a baseline operating state and deeply understand the mechanics of the tools you rely on. Which leaves you with one final thought to ponder as we wrap up. We mentioned DeepSeek offering near frontier AI capabilities for just 60 cents a million tokens, while giants are committing
hundreds of billions of dollars to physical hardware. It raises a massive question about the macroeconomic reality of tech right now. If the actual commercial value generated by these AI models is being aggressively commoditized and replicated by open source alternatives for literal pennies, are we staring directly at the imminent burst of an AI hardware bubble? Because right now, the astronomical cost of building the infrastructure seems to be vastly outpacing the revenue those models can ever mathematically hope to generate. A truly sobering thought to
end on, and something every tech leader needs to factor into their long -term strategy. Absolutely. Thanks for joining us for this deep dive into the underlying mechanics of our digital world. Keep a close eye on your perimeter, test your patches before you push them, and we will see 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;...