Patched and Still Broken: Zyxel Switches, Apple Zero-Days, and Why Your AI Tools Are Leaking
Three stories landed this week that individually would be worth your attention. Together, they describe something more uncomfortable: the routine failure of security measures that businesses are actively relying on.
A network switch patch that doesn’t stop the attack it claims to fix. A patched Apple vulnerability that CISA has confirmed is being exploited in the wild, now tied to Russian state actors targeting mobile users at scale. And a vulnerability in ChatGPT’s highest-security mode that was silently leaking conversation data without any action from the user.
If you run Zyxel GS1900 switches, use Apple devices, or have staff using AI tools with business information, read on. This is not abstract.
Zyxel GS1900: The Patch That Doesn’t Patch
Zyxel’s GS1900 series network switches are common kit in small and medium-sized business networks. They are affordable, capable managed switches, and many UK businesses have them sitting quietly in a server room or comms cabinet, largely unthought-of.
This week, researchers at Binarly published findings showing that Zyxel’s earlier fix for an actively exploited flaw in the GS1900 line does not actually block the exploit. The vulnerability in question allows unauthenticated root command execution: an attacker with network access to the device can run commands at the highest privilege level, with no login required.
Zyxel has now issued a further fix. But the sequence here matters. A vendor issued a patch. The patch was applied. The device appeared protected. It was not.
For a small business owner or their IT support, this represents a specific and underappreciated failure mode. The assumption that applying a vendor patch closes a vulnerability is baked into most patch management processes. In this case, that assumption would have left your network switches fully exploitable, regardless of whether you had done everything right.
The two CVEs involved are CVE-2026-107441 and CVE-2026-7273. If you have Zyxel GS1900 switches in your environment, the action is not to check whether the first patch was applied. The action is to verify you have the latest firmware from Zyxel’s support pages and confirm it post-dates Binarly’s disclosure.
If your managed service provider handles your network hardware, ask them to confirm the current firmware version and when it was last verified. Not updated. Verified.
Apple CVE-2024-27834: Actively Exploited, Linked to APT29
On the 5th of November 2026, CISA added CVE-2024-27834 to its Known Exploited Vulnerabilities catalogue. The flaw is a pointer authentication bypass affecting multiple Apple operating systems. CISA’s addition to the KEV catalogue means the agency has confirmed evidence of active exploitation in the wild, not theoretical risk.
The same vulnerability has been linked to activity by APT29, the threat group attributed to Russia’s SVR foreign intelligence service. Reporting from Dark Reading describes APT29 compromising a Ukrainian SEO and website analytics service and using it as a watering hole: a legitimate site, modified to serve exploits to visitors. The targets were mobile users, and the scale is described as thousands of individuals.
A watering hole attack works by compromising a website that the target population is likely to visit. The attacker does not need to reach victims directly. They wait for victims to come to them, through a trusted site the victim has no reason to suspect.
For UK small businesses, the immediate relevance is straightforward: if your staff use iPhones, iPads, or Macs, and those devices have not been updated recently, CVE-2024-27834 represents a confirmed, actively exploited exposure. The patch exists. The question is whether it has been applied.
Apple pushes updates through standard software update mechanisms on all its platforms. For iPhones and iPads, that is Settings, then General, then Software Update. For Macs, it is System Settings, then General, then Software Update. The fix is available. The gap is in whether it has been applied.
If your business manages a fleet of Apple devices, verify update status across the fleet. If staff manage their own devices, send a direct communication today asking them to check and update. Do not assume it has been done.
ChatGPT Lockdown Mode: Silent Data Exfiltration Without User Interaction
The third story is the one that should concern any business where staff are using AI tools in the course of their work, which at this point is most businesses.
Researchers disclosed a now-patched vulnerability in ChatGPT (CVE-2025-51591) that allowed an attacker to bypass Lockdown Mode and exfiltrate conversation data without any interaction from the user. Lockdown Mode was introduced precisely to restrict external data connections and protect sensitive conversations. The vulnerability used a flaw in ChatGPT’s HTML preview functionality to cause crafted content to escape those restrictions.
The practical implication: a user operating in what they believed was a hardened, restricted security mode could have their conversation content silently read and transmitted externally, without clicking anything, approving anything, or receiving any visible indication that anything had happened.
OpenAI has patched the vulnerability. But the disclosure raises a question that the patch does not answer: what are your staff sharing with AI tools, and under what assumption of confidentiality?
If an employee pastes client data, financial figures, internal strategy, or personal information into a ChatGPT conversation, that content is not on your infrastructure. It is on OpenAI’s. Any vulnerability in that platform is a vulnerability in your data handling, regardless of what your internal security controls look like.
This is not an argument against using AI tools. It is an argument for having an explicit policy about what category of information is acceptable to share with them, and enforcing it.
The Pattern Underneath All Three
Take a step back from the individual incidents and the common thread becomes visible.
In each case, a security control that businesses and individuals were actively relying on failed to provide the protection it appeared to offer. A patch that did not close the vulnerability. A platform security mode that could be bypassed silently. A device security assumption undermined by active exploitation of a known flaw.
The failure is not in any single vendor or product. The failure is in the assumption that declared security equals actual security. It frequently does not.
For a small business without dedicated security staff, this is genuinely difficult. You cannot verify every vendor claim. You cannot audit every patch. But you can build verification into the questions you ask your IT support or MSP, and you can maintain a sceptical baseline rather than a trusting one.
How to Turn This Into a Competitive Advantage
Businesses that treat security verification as routine rather than exceptional are increasingly distinguishing themselves during procurement and due diligence processes.
If you can demonstrate to a prospective client or partner that your patch management includes post-application verification, that your device update status is monitored rather than assumed, and that you have a documented policy on AI tool usage and data classification, you are ahead of most businesses of your size.
These are not expensive capabilities. They are process disciplines. And they are increasingly the questions that larger organisations ask of their supply chain before signing contracts.
Making the Business Case
Three points for the conversation with whoever controls the budget or sets policy:
First: Applying a patch and verifying a patch are different activities. The Zyxel case illustrates that a device can appear protected while remaining fully vulnerable. The cost of verification is a few minutes per device. The cost of unverified exposure is potentially the entire network.
Second: State-sponsored actors are actively exploiting consumer-grade vulnerabilities on devices your staff carry in their pockets. APT29 targeting Apple devices via watering hole attacks is not a nation-state problem. It is a device management problem. The fix is an update. The question is whether you know if it has been applied.
Third: Staff use of AI tools is a data governance question, not just an IT question. If you do not have a policy on what can be shared with external AI platforms, you are making an implicit decision to have no controls. That implicit decision has consequences under GDPR if personal data is involved.
What to Do Before the End of This Week
Check your Zyxel GS1900 firmware. If you have these switches, do not assume the earlier patch is sufficient. Verify the current firmware version against Zyxel’s latest advisory and ensure it post-dates the Binarly disclosure from the 5th of November 2026. If your MSP manages your network hardware, ask them for written confirmation of the current firmware version on each affected device.
Audit Apple device update status. For every iPhone, iPad, and Mac used for business purposes, confirm that the latest available software update has been applied. For company-managed devices, run this through your mobile device management platform. For personal devices used for work, send a direct request to staff today.
Review your AI tool usage policy. If you do not have one, write one this week. It does not need to be long. It needs to answer two questions: what categories of information are acceptable to share with external AI tools, and what is not. Client data, financial data, personal data, and internal strategic information should sit in the second category. Circulate the policy to all staff and confirm receipt.
Ask your IT support to demonstrate, not just assert. For any patch or security control you rely on, ask your provider to show you the verification evidence, not just confirm the action was taken. This is a reasonable question. If your provider cannot answer it, that is information you need.
The research behind this episode is linked in the sources below. The Binarly findings on Zyxel, CISA’s KEV catalogue entry for CVE-2024-27834, and the ChatGPT disclosure are all worth reading in full if you manage security decisions for your business.
If this was useful, follow the show wherever you listen, leave a rating or a review, and drop a comment with what you would like covered next. If you know someone running a small business who should be hearing this, send it to them. It takes thirty seconds and it might save them considerably more than that.