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 - Citrix's Shut-It-Down Weekend, a Mail Gateway With No Patch, and the AI Price War Arrives

Oct 4, 202622 minAudio
ShareXLinkedInFacebookWhatsAppEmail

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 5 October 2026, covering the week from 27 September. Citrix NetScaler zero-days (27 Sep). CVE-2026-88771 (unauthenticated command execution, all deployments including default config) and CVE-2026-88772 (memory overflow, RCE when DTLS is on - the VPN default). Citrix: CVSS v4.0 9.5 for both. Fixed in 14.1-73.37, 13.1-64.23 and FIPS/NDcPP builds; 12.1 and 13.0 are end-of-life with no patch. CISA KEV 27 Sep, deadline 30 Sep. Mandiant: exploitation since at least 3 Sep, "dozens" of victims across government, finance, technology, education and professional services; suspected state-sponsored actors using the WHIPSHOT web shell and SLAPSHOT tunneller. Public PoC 28-29 Sep. ASD's ACSC alert 28 Sep, later updated: Australian organisations have confirmed exploitation; hunt back to 4 Sep. Patching does not remove existing web shells. Cisco Catalyst SD-WAN Manager (30 Sep). CVE-2026-76504, unauthenticated API authentication bypass to admin, CVSS v3.1 9.8 per Cisco, exploited in the wild, no workaround. Fixed in 20.9.10.1, 20.12.8.2, 20.15.6.1, 20.18.4.1, 26.1.2.1 and 26.2.1. CISA KEV 30 Sep. FortiMail zero-day (1 Oct). CVE-2026-104286, unauthenticated arbitrary file write via path traversal in the web interface's identity-based encryption (IBE) component, CVSS v3 9.8 per Fortinet, exploited in the wild. Affects 8.0.0-8.0.1, 7.6.0-7.6.6, 7.4.0-7.4.8, 7.2.0-7.2.9. Fixes 8.0.2, 7.6.7 and 7.4.9 were still "upcoming" at 3 Oct; 7.2 must migrate. Fortinet's workarounds: disable IBE, or block internet access to the FortiMail webmail interface. CISA KEV 1 Oct, deadline 4 Oct. Apple CoreGraphics zero-day (28 Sep). CVE-2026-86950, out-of-bounds write; fixed in iOS/iPadOS 26.7.1, macOS Tahoe 26.7.1, macOS Sequoia 15.8.1. Apple: exploited in "an extremely sophisticated attack against specific targeted individuals" on iOS before 27. Reported by Meta Product Security. Crash PoC (PDF with crafted font) published 30 Sep (US). No vendor CVSS score. Microsoft Digital Defense Report 2026 (1 Oct). Median time from in-the-wild discovery to weaponisation "well below 24 hours"; AI used for personalised phishing, custom malware and faster data theft; government the most-targeted sector at 27%. AI price war. OpenAI DevDay (29 Sep): GPT-6.1 Sol at about one-fifth of flagship token prices (US$2/US$10 per million input/output); always-on "Dots" agents (off by default for enterprise); ChatGPT in Slack and Teams; Codex Security Cloud; Pro 500 at US$500/month. Google Gemini 4 Argon (30 Sep US): vetted cyber defenders first, broad release to follow; introductory US$2/US$10 per million tokens. Anthropic draft prospectus (Reuters, 28 Sep). Reported 2025 revenue ~US$4.6bn, net loss ~US$42bn (mostly non-cash), compute costs US$7.33bn, US$518bn infrastructure commitments; valuation target above US$2tn. Draft figures; Anthropic declined to comment. Disclosure: this podcast is produced with help from an Anthropic model. OFX Group (2 Oct). ASX-listed payments company investigating unauthorised access to data including some client and job-applicant data; no access to accounts or funds identified; ACSC, OAIC and overseas regulators notified; client numbers not yet known. Coming up: Wednesday, Stake and the DriveWealth breach. Friday, Fakturownia - 600,000 businesses' invoicing data. Visit www.kinsoft.com.au to talk through your security and IT needs. Sources: Citrix CTX697096; CISA KEV; Mandiant; ASD's ACSC; BleepingComputer; Help Net Security; CyberScoop; Tenable; Cisco; Rapid7; Fortinet FG-IR-26-175; The Hacker News; Microsoft; OpenAI; Google; VentureBeat; Reuters; Fortune; OFX ASX announcement.

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 waking up, right, on a Sunday morning. The American cyber agency, CISA, has just issued an emergency directive. Oh, which is never a good way to start a Sunday. Exactly. And they give your IT team exactly three days to patch a critical vulnerability in your network's front door. So, you know, you scramble, you apply the patch, and you sit back thinking, well, your data's safe. Yeah, you think you did your job. Right. But what you don't realize is that... state -sponsored attackers actually used an automated

exploit to breach your systems like three weeks ago. So you aren't actually securing your network. You're basically just locking the door behind the people who have already set up a camp in your server room. I mean, it completely breaks our traditional understanding of incident response. The whole architecture of enterprise security is built on this grace period that simply does not exist anymore. Welcome to the collapse of digital time. That is actually our mission for this deep dive. We are unpacking a pretty dense stack of tech industry briefings from early October

2026. We're pulling really heavily from Kinsoft's last week in tech. It's a great source. It really is. And we're using it to understand how the timeline of cybersecurity has been, well, effectively shattered. We're looking at the servers holding your data, the phone in your pocket, and, of course, the artificial intelligence engines driving this relentless acceleration. Yeah, the AI piece is massive. Right. And here's where it gets really interesting. Microsoft just released their new digital defense report, and they put a highly

specific number on this collapse. Yeah, this number is wild. Based on their telemetry, the median time between a digital flaw being discovered in the wild and an attacker weaponizing it is now, quote, well below 24 hours. Which means the traditional monthly patch cycle is, well, it's officially a relic of a bygone era. Basically dead. Yeah. The whole concept of, you know, gathering a list of software updates, running them through a staging environment for a couple of weeks and then rolling them out during a quiet maintenance

window. Right. That was built for human speed threats. When the time to exploitation drops below 24 hours, you aren't managing risk anymore. You are literally actively bleeding. Wow. So if edge devices like net scalers are already compromised before a patch even drops, the corporate perimeter is essentially dead. Completely. Let's look at what this 24 -hour metric actually looks like in practice, because for IT administrators, this translates into absolute chaos. Let's break

down the Citrix Netscaler crisis from late September. Oh, man, what a nightmare that was. Right. For anyone who hasn't managed one, a Netscaler is a critical edge device. It sits outside the firewall, and it basically acts as a load balancer and a remote access portal for corporate VPNs and internal applications. Yeah, they are massive targets because they are literally designed to be exposed to the public Internet. Right. And furthermore, they operate as black boxes. You know, security teams can't easily just install

standard endpoint detection and response software on a proprietary appliance like a net scaler. So they can't see what's happening inside it. Exactly. So when they get compromised, the attackers have this massive blind spot to hide in. And hide they did. I mean, the timeline to this incident is just wild. Rumors started swirling on forums on Friday, September 25th. Admins were saying they were being quietly advised by vendors to just shut their net scalers down entirely. Not patch them, shut them off. Which is nuclear.

Literally. Then on Sunday, September 27th, Citrix drops a bulletin detailing eight vulnerabilities with two of them already being exploited to zero days. We were talking CVSS 9 .5 out of 10. Yeah, basically as bad as it gets. Right. The first one, CVE -2026 -88771, is an input validation flaw. Now, I understand input validation issues on a web app, but how does an enterprise appliance just allow an unauthenticated attacker to run commands directly? Well, it comes down to how

these legacy management interfaces process incoming requests. If the administrative interface doesn't strictly sanitize the parameters of an incoming HTTP request, an attacker can craft a payload that breaks out of the expected data format. Oh, I see. Right. And then it injects arbitrary system commands into the underlying operating system. Because these appliances run with high privileges, you know, to manage network traffic, that injected command executes with root -level authority. Wow. And that affected the default configuration for everyone. Yep. Then the second

critical flaw, CVE -2026 -88772, that was a memory overflow vulnerability. And that one required a protocol called DTLS to be turned on. Right, which is Datagram Transport Layer Security. It's a protocol used to speed up VPN connections. Okay. And crucially, it is enabled by default on VPN virtual servers. Of course it is. Yeah. So if an attacker sends a specially crafted packet that's larger than the memory buffer allocated to process that DTLS handshake, the excess data

overflows into adjacent memory spaces and it overwrites executable code. So on that Sunday, CISA gives federal agencies three days to patch. But here's the chronological nightmare we mentioned earlier. Mandiant, Google's incident response team, reported that exploitation of that DTLS flaw actually began on September 3rd. Weeks earlier. More than three weeks before Citrix even had a patch, attackers had already deployed custom persistence tools, specifically a web shell called Whipshot and a tunneling tool called Slapshot.

Which brings up a really critical question. If you apply the Citrix patch on September 28th, does that patch wipe out Whipshot and Slapshot? What's fascinating here is, well, no, not at all. And this is a fundamental misunderstanding of what a patch actually does. A patch only modifies the specific vulnerable code. So in this case, it's just fixing the memory buffer allocation for the DTLS handshake. It's not antivirus. Exactly. It does not scan the appliance's file system for unauthorized scripts. Once an attacker uses the memory overflow to drop a web shell like

Whipshot into a directory that the web server reads, they have a permanent backdoor. Wow. The Australian Cybersecurity Center had to issue very explicit guidance on this. They warned organizations that patching was insufficient. Administrators had to manually hunt through their system logs, going back to at least September 4th, looking for the specific file creation events. Yeah, and anomalous outbound traffic patterns associated with those tools. So if Mandiant says exploitation began on September 3rd, but the patch wasn't available until September 27th? I mean, isn't

applying the patch just like changing the locks after a burglar has already moved into your guest bedroom? That is exactly what it is. If you just patched and walked away, you were essentially changing the locks on the front door while the attackers were sitting in your living room, routing your database out through a slap shot tunnel. And the attackers know this window is shrinking, which is why they aren't stopping at just one vendor. Right. This Netscaler crisis was just the first wave. Right on its heels, we saw a massive CVSS 9 .8 flaw in Cisco's Catalyst SD -WAN manager. That's CVE -2026 -76504. Yeah.

And this is the central console used to manage networks across multiple physical locations, right? Right. Another highly privileged internet -facing edge device. And the mechanism here was an API authentication bypass. Right. How does an attacker bypass authentication on something as critical as an SD -WAN manager? Well, usually it involves manipulating the API's token validation logic. If the system fails to properly verify the cryptographic signature of an authentication token, or if it mishandles a specific string

of characters in the HTTP header, an attacker can trick the API into granting their request the same privileges as a legitimate authenticated administrator. Oh, man. From there, they can just... push malicious routing configurations to every physical branch office connected to that manager. It's just devastating. But the vulnerability that really captured the sheer panic of the speed was Fortinet. Oh, the FortiMail issue? Yeah. On October 1st, they disclosed CVE -2026 -104286, another 9 .8 severity flaw in

FortiMail, which is their email security gateway. The underlying issue was path traversal, but... Instead of just listing that, let's unpack the mechanics. How does a path traversal flaw on a mail gateway translate to attackers reading corporate emails? So a path traversal vulnerability means the software can be tricked into reading or writing files outside of its restricted intended directory. Okay, so it gets out of its sandbox. Exactly. In the context of a mail gateway, the web interface is supposed to only access specific

web assets. But... If an attacker sends a request with manipulated file paths, you know, using sequences like dot dot slash, they can navigate the file system to reach the raw mail spools or the configuration files containing encryption keys. And the briefing noted they were actively using this to plant files and alter the configuration. So the system would quietly forward sensitive corporate emails to attacker controlled servers. But here is the terrifying part about the 40 mail situation. Attackers were exploiting this

in the wild. But when IT admins went to Fortinet's site to download the fix, versions like 8 .0 .2 and 7 .6 .7 were just listed as upcoming. That is the ultimate nightmare scenario. You have active exploitation happening on your perimeter, and the vendor simply hasn't finished compiling the patch. And it gets worse for legacy infrastructure. If you were running their older 7 .2 branch, the vendor indicated you weren't going to get a fix at all. So you had to completely migrate your entire mail security infrastructure to a

newer branch while actively under attack. Yeah, good luck with that. When the patch doesn't exist, all you have are architectural amputations. You have to physically disable features or completely sever the 40 -mail interface from the public internet. Which is... effectively blinding your own remote workforce just to stop the bleeding. Which perfectly illustrates the structural shift we're discussing. Attackers are finding and weaponizing these flaws so quickly that the traditional defense mechanism, you know, vendor patches, is fundamentally broken. And this dynamic isn't just a problem

for massive corporate server rooms. It is targeting the device sitting in your pocket right now. Absolutely. Let's talk about the Apple Zero Day that hit that same week. Because it brings this chronological collapse right to you. Apple had to push out emergency patches across the board. iOS and iPadOS 26 .7 .1 and macOS Sequoia 15 .8 .1. Yeah. They were fixing CVE -2026 -86950 and out of bounds right in a component called Core Graphics. Okay, let's unpack this. I understand

servers getting attacked, but Core Graphics is a visual rendering engine. Right. How does a visual engine translate to remote code execution? Well, if we connect this to the bigger picture, it seems counterintuitive that drawing a picture on a screen can compromise a device. But rendering engines process incredibly complex, highly structured binary formats, things like PDFs, SVGs, and custom font files. To display a font, core graphics has to parse thousands of rules about kerning, vector paths, and ligatures. So it's not just

displaying a picture. It's actually executing a set of complex instructions embedded in the file. Exactly. Think of your system's memory like a strict spreadsheet. The core graphics software is only given permission to write its rendering data into column A. But a maliciously crafted font file... acts like a corrupted macro. It deliberately feeds the engine malformed instructions that cause it to spill over, writing outside of column A and overwriting the critical system commands sitting in column B. Right. So the attacker

essentially rewrites the rules of the operating system on the fly just by making the phone try to display a corrupt font. That is a highly accurate way to visualize an out -of -bounds write. And notice the delivery mechanism. A user doesn't have to navigate to a shady website or grant a sketchy app permissions. You don't do anything. They just have to receive a file that forces the operating system to render that font. It's crazy. Furthermore, we have to look at who reported this vulnerability. It wasn't found by Apple's internal bug bounty team. It was reported by

Meta's product security team. And Apple explicitly noted it may have been exploited in highly targeted attacks. And when tech giants say highly targeted. They usually mean state -sponsored surveillance or corporate espionage. Exactly. The average person listening to this deep dive might not be the primary victim today, but if you manage traveling executives or lawyers carrying sensitive client data, they are absolutely in the crosshairs. Yeah, 100%. By the time that PDF hits a device, the exploitation has already happened. Relying

on users to click update is a failed strategy. Proactive, enforced mobile device management is the only way to operate when the attack window is zero. Yeah. Which brings us to the core driver of all this. How are attackers compressing these timelines from months to days and now down to mere minutes? Right. The answer is artificial intelligence. Going back to that Microsoft 2026 digital defense report, the data paints a very clear picture. State -sponsored attackers, specifically groups operating out of China, Russia, and North

Korea, are actively deploying large language models in their offensive operations. It's important to clarify Microsoft's caveat here, though. We are not yet seeing fully autonomous, self -directing AI systems launching cyber wars on their own. So Skynet isn't here quite yet. Right, not yet. The strategic objectives are still set by human operators. However, AI is acting as an unprecedented force multiplier for the tactical execution. They're using AI to parse vulnerability disclosures the literal second they hit the internet. Automatically

generating customized exploit scripts for flaws like those net scaler or 40 mail bugs. They're using LLMs to ingest a target company's public communications and instantly generate hyper -personalized spear phishing campaigns. That's alarming. Microsoft noted that the government sector jumped from being the target of 17 % of observed activity up to 27%. That is a huge jump. A massive spike. entirely facilitated by the fact that AI removes

the friction of manual reconnaissance and manual coding. Think about your own company's defenses for a moment. Are they architected to stop a human typing on a keyboard, constrained by human typing speed and human sleep cycles? Or are your defenses built to stop an automated engine generating thousands of unique polymorphic malware variants every minute? Because that automated engine is getting cheaper and faster by the day. The exact same AI technology empowering these rapid cyber attacks is currently the subject of a ruthless

corporate arms race. Oh, yeah. The cost of advanced intelligence is plummeting, while the capabilities are just being pushed to the extreme. Let's look at that arms race, because the back -to -back product announcements from OpenAI and Google in early October were staggering. Just massive news. OpenAI held their developer conference and announced GPT 6 .1 Sol. They are pitching this model as hitting near their flagship performance metrics, but for a fifth of the price. On top

of that, they are natively integrating ChatGPT into Slack and Microsoft Teams, launching a $500 a month pro tier and rolling out something called Dots. And the Dots feature is fundamentally changing the attack surface of the modern enterprise. Because a dot isn't just a chatbot, right? It's an always -on agent. OpenAI gives it a dedicated cloud computer, its own web browser, and it just keeps working in the background when you log off. It's available on their Consumer Pro Plan and Business Premium. Right. So what does this

all mean? If an AI agent keeps working after I log off and it has access to my Slack, isn't that a massive governance nightmare waiting to happen? I mean, who is liable if it gets manipulated? It is the most critical governance policy any business needs to draft this week. You have to map out exactly who in your organization is authorized to switch on an autonomous agent. Because if the answer is anyone with a corporate credit card, your network is functionally compromised by design. You are effectively hiring invisible,

unmonitored employees who have read and write access to proprietary data and interact with the outside Internet 24 -7. It completely breaks traditional identity and access management. And Google didn't even wait a full 24 hours to fire back. The very next day, they announced Gemini 4 Argon. It's leading most industry benchmarks, and their introductory pricing is $2 per million input tokens and $10 per million output tokens, which exactly matches OpenAI's aggressive new pricing model. Yep, down to the dollar. But here

is the most fascinating part of Google's strategy. They aren't releasing Argon to the general public yet. They are giving it first to a vetted program of trusted cyber defenders. And they are providing a version with the standard cyber safety guardrails completely removed. That is a profound acknowledgement of the threat landscape we just mapped out. Google recognizes that the only way to fight AI speed attacks is with unfiltered AI speed defense. Fight fire with fire. Exactly. Defenders need raw models that can analyze malicious code and

generate countermeasures without tripping over content moderation filters that were, you know, designed for consumers. Right. But to understand why this price war. $2 a million tokens is so fierce, we have to look at the underlying economics of what it takes to run these frontier models. Reuters reported on a leaked draft of Anthropic's IPO prospectus. Oh, that leak was eye -opening. Now, a draft is a draft, but the leaked figures outline the brutal reality of the AI industry. They project revenue jumping to nearly $4 .6

billion in 2025. But they also project a net loss of around $4 .2 billion. Right. Now, a massive chunk of that $42 billion is a non -cash accounting charge related to stock -based compensation and warrants. Okay, fair. But even if you strip that out, their operating loss was still projected above $8 billion. And the most critical number in that entire leaked document, compute and infrastructure costs, were $7 .3 billion. Wait, so over half of all their operating expenses just go to keeping the servers running and the microchips crunching

the data? Yes. Compute is the actual battlefield. The $2 token price from Google and OpenAI is heavily subsidized. They are absorbing those astronomical compute costs to starve competitors and capture the enterprise layer. Makes sense. Eventually, the subscriptions your business pays for AI will have to rise to cover that true infrastructural cost. But right now, this artificial price depression is making advanced AI capabilities highly accessible to everyone, including threat actors. It is incredibly easy to get lost in the sheer scale of $7 billion

compute bills, CVSS 9 .8 vulnerability scores, and sub -24 -hour weaponization metrics. But we have to ground all this high -level architecture in immediate human impact. Right. What happens when these ultra -fast attacks actually succeed? Well, we can look at the OFX group breach that occurred right in the middle of this chaotic week. OFX is a major international payments company based in Sydney. They had to notify the Australian Securities Exchange that they were investigating a cyber incident involving unauthorized access

to data. But notice who was affected. It wasn't just active clients. It was people who had dealt with OFX in tangential ways, specifically job applicants. Right. OFX stated their core services were running normally. They hadn't identified any access to client funds. And the stolen data couldn't be used to initiate payments. Which is the good news. That is the good news. But they still stole the applicant database. Why target job applicants? Because an applicant database contains pristine, highly structured personal

data. Names, private emails, detailed work histories, direct phone numbers. Oh, of course. In an era where AI can generate hyper -personalized spear phishing campaigns in seconds, that raw data is incredibly valuable ammunition. Attackers feed those resumes into an LLM and the model instantly drafts perfectly contextualized, highly convincing emails targeting specific employees at other companies. Personal data is the fuel for the next wave of automated attacks. Which brings us full circle. We are operating in a

digital landscape where the fundamental rules of time have just been shattered. We have patches for edge devices like Citrix, Netscalers, and Fortinet gateways that are obsolete weeks before they were even compiled. We have endpoint vulnerabilities rewriting memory rules on the fly. We have state -sponsored attackers using artificial intelligence to compress the window of exploitation down to minutes. Yep. And simultaneously, a cutthroat corporate price war is subsidizing the cost of this intelligence, putting powerful autonomous agents into the hands of anyone with a credit

card. This raises an important question, though. If state -sponsored attackers are using AI to compress attack times to mere minutes and businesses are deploying always -on AI agents like OpenAI's DOTS to automate their networks, are we rapidly approaching a point where human reaction time is completely obsolete? That's a scary thought. Will the future of cybersecurity just be our AI agents fighting their AI agents in the dark while we humans just sit back and wait to see who won? It really makes you rethink what it means to be fast enough, doesn't it? If the attackers

are already mapping your network before CISA even issues the warning, changing the locks isn't going to save you. You need a completely different architecture to secure the house. Thank you so much for joining us as we unpack this incredible collapse of time in the digital world. Keep questioning your assumptions, keep a very close eye on your edge devices, and we'll catch you on the next deep dive.

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