Cyber, in brief

Microsoft released fixes for 1,411 flaws in its software on Tuesday, the monthly day it sends out security updates. [1]

That is the most it has ever fixed in one month. The record before was 974, set only in September. [1][3]

Counts differ by how they are made: BleepingComputer says 1,411, Dark Reading 1,413, and Brian Krebs, a security reporter, more than 1,400. [1][2][3]

Attackers were already using one of the flaws before the fix. It is in Windows Shell, the part of Windows that opens files and folders. [1][3]

Windows tags every file that comes from the internet with a label called the Mark of the Web. The flaw lets an attacker send a file that arrives without that label. [3]

At least five Windows and Office safety checks read that label, including the warning before a downloaded program runs and the block on Office macros, small programs inside documents. Without the label, none of them stops the file. [3]

Dark Reading reports that Claude, the AI made by Anthropic, found this flaw. [3] Microsoft has not said how attackers used it. [1]

CISA, the US government's cyber-defence agency, added the flaw to its list of flaws known to be used in attacks on Tuesday. [4]

Two other flaws were made public before the fix, one of them in .NET, a toolkit many Windows programs are built on. [1] Microsoft says nobody is known to be using the .NET flaw yet. [2]

Microsoft says AI tools found many of this month's flaws. It has also started grouping them into 15 sets, to help companies choose what to install first. [2][1]

Tyler Reguly of Fortra, a security company, says only 26 Microsoft flaws this year have reached CISA's list of flaws used in attacks. That is under 1% of all Microsoft fixed. [3]

On home computers the update arrives through Settings, then Windows Update. [7] Windows 10 machines in Microsoft's extended-update scheme get the same fixes. [6] Windows 11 version 24H2 Home and Pro got its last update on Tuesday, so those machines need a newer version. [5]

Why it matters. Watch CISA's list in the coming weeks: it will show whether attackers start using more of the 1,411 flaws now that the fixes are public. [4][3]

Ivanti told customers on Tuesday to update its Connect Secure boxes at once. These boxes are VPNs, which let staff log in to a work network from home. [9] Volexity, the security firm that found the attack, says attackers chained three flaws to take full control of boxes that were fully updated. [11] Ivanti has fixed two, and the third is still open. [11] Ivanti says it found part of the flaw itself in June, but held the fix for its normal release because nobody was using it yet. [10]

Why it matters. CISA says Ivanti's own checking tool may miss this attack, because the commands run in memory and leave no files behind. [12] It tells anyone who suspects a break-in to reset the box and change every password on the network. [12]

The US Justice Department has charged Abdulrahman Alfozan, an engineer at OpenAI, the company behind ChatGPT. Prosecutors say he hid a recording device in an executive meeting room at its San Francisco office. [13] OpenAI staff found the device on 30 September, wrapped in toilet paper and plugged in behind a television. [13] Prosecutors say a chatbot account linked to his wife asked how to build and hide a recorder. [13] He was arrested on 9 October, and OpenAI fired him. [13][14]

Why it matters. OpenAI says it has no sign that he took its AI models, code or customer data. [14] Neither prosecutors nor OpenAI have given a motive. [15] The FBI director, Kash Patel, says the aim was to capture talk about OpenAI's AI work. [16]

The US Justice Department charged 14 people on Tuesday over a scheme that put North Korean workers in remote IT jobs at US companies. [18] In September police searched more than 40 Texas homes thought to be laptop farms. [17][18] Helpers there keep company laptops switched on so the workers seem to be in the US. [18] They seized 187 laptops and other devices. [17] One group alone earned about $3 million from about 70 companies in 17 states. [17]

Why it matters. The US government says the wages pay for North Korea's weapons programmes. [17] The FBI says the workers now pay Americans to sit job interviews for them, and it wants checks to continue throughout a job, not only at hiring. [18]

France's finance minister, Roland Lescure, told parliament on Tuesday that about 700 of the country's 10,000 supermarkets were hit by the attack on their fridges. [23] BFM-TV, a French news channel, reports that hackers set the fridges as warm as 50C. [23] Les Mousquetaires, which runs Intermarche, says 300 of its shops were hit, and Cooperative U, another French grocery chain, says 80. [23] France's cyber agency says control boxes made by Copeland, a US firm, were broken into, as well as those of Danfoss, a Danish company. [23]

Why it matters. Danfoss says most of the boxes hit were older models that no longer get security updates. [23] The minister says France's food supply is not at risk, while the Paris prosecutor investigates the spoiled food and reported fires. [23]

Ahold Delhaize, one of the world's biggest supermarket groups, said on Tuesday it had begun restarting systems it switched off after a security problem. [24] It owns Albert Heijn in the Netherlands, Delhaize in Belgium, and Stop & Shop, Food Lion and Giant in the US. [23] Its shops stayed open, but the websites of those chains were still offline on Tuesday. [23] Local media reported that deliveries and online orders were hit in Belgium and the Netherlands. [24]

Why it matters. Ahold Delhaize says it has no sign that customer or staff data was stolen. [23] Nobody has said whether its problem is linked to the attack on French supermarket fridges. [24]

TFG London, the UK arm of South Africa's Foschini Group, owns the fashion chains Phase Eight, Hobbs and Whistles, and it found a cyber attack on 11 October. [26] Several of its IT systems were still down two days later, though its shops stayed open. [26][27] It says payments are handled by another company and it has no reason to think card details are at risk. [27] It is still checking whether customer records were taken, and has told the UK's data watchdog. [27]

Why it matters. No gang has claimed the attack yet. [26] H&M, the Swedish clothing group, is separately checking a claim online that hackers broke into it. [28][25]

Police in Ghana and Nigeria arrested or held 28 people in raids on scam centres, US and British officials said on Tuesday. [19] In one compound near Accra, Ghana's capital, police found 117 people believed to have been trafficked, people brought there by force or by tricks, and made to run scams. [19] Fifteen police agencies and the same number of tech companies, including Meta and Google, took part. [19] Meta says it removed more than 3.7 million accounts linked to the centres. [20]

Why it matters. The scams included fake police calls and fake crypto investments aimed at people around the world. [19] Officials found more than $11 million that could be frozen, but did not say how much they actually froze. [19]

The FTC, the US consumer watchdog, says Digital Forensics Corporation, an Ohio firm, took at least $60 million from people being blackmailed online with intimate photos. [21] Victims paid $1,499 to $20,000 for help tracking the blackmailer and calling the police. [21] The FTC says the firm did none of it, and faked chat logs and locations to look busy. [21] A court has ordered it to stop trading while the case goes on. [21]

Why it matters. The FBI tells such victims to ignore and block the blackmailer, but the firm told them that was the last thing to do, the complaint says. [21] Its sites were advertised on Google, YouTube and X to people searching for help. [22]

Docker has fixed a critical flaw in Docker Hub, an online library where programmers share ready-made software packages called images. [29] An attacker could make an account, turn it into an organisation and name a victim as its owner. [29] Docker Hub then handed over a login key that let the attacker publish images under the victim's real name. [29] The flaw was there from August 2024, and Docker fixed it within four hours of a report on 30 September. [29]

Why it matters. Anyone downloading an image trusts the name on it, which is why Docker rated the flaw critical. [29] It says it checked 304,000 uploads made with the handed-over keys and found no sign of misuse. [29]

Meta, which owns Facebook and Instagram, is writing to about 1,500 people, mostly its own US staff. [36] Hackers got into some of its code stores on GitHub, a site where programmers keep their code. [36] They got in with one employee's stolen login. [36] Meta says 99% of what they took was code, and 1% was personal data that should never have been stored there. [36] Some of that data was made-up test data, so the hackers may not know which records are real. [36]

Why it matters. Meta says no Meta product had a flaw and its internal systems were not breached. [36] It is adding scans to catch personal data put on GitHub by mistake. [36]

SAP, the German company whose software runs many firms' accounts and stock, fixed 20 flaws on Tuesday. [30] The worst, in a toolkit for building AI agents and apps on SAP's cloud, gets the maximum danger rating. [30] An attacker with no login could redirect the app's sign-in requests to their own server and steal its secret keys. [30] SAP also fixed two critical flaws in other products. [30]

Why it matters. Companies that cannot update straight away can block the attack by only accepting a list of known customer names. [30] An SAP flaw fixed in August was used by attackers days after its fix. [30]

Wordfence, a company that protects WordPress websites, found a serious flaw in Ghostscript, free software that turns PDFs into pictures for previews. [31] A specially made PDF could let an attacker run their own code on the server that makes the preview. [31] Ten common versions of Linux carried the faulty code. [31] Artifex, which looks after Ghostscript, released a fix on 8 September, and Wordfence has not seen anyone use the flaw. [31]

Why it matters. Any website that previews PDFs people upload can be at risk, not only WordPress. [31] Site owners are told to check with their host that Ghostscript is version 10.08.0 or later. [31]

Anthropic, the company that makes Claude, says small AI programs may be months from copying themselves across cheap internet-connected gadgets, like TV boxes and smart ovens. [32] Its researchers split one AI model across 15 TV boxes bought for about $55 each. [33] In a test network of eight pretend Android devices, one model spread to all eight in four of five tries. [33] They did not show a working self-spreading AI on real devices. [32]

Why it matters. Anthropic says the skills that make AI good at writing code also carry over to hacking. [35] Such a program would have no central point to switch off, it says. [32] The Register calls it fear about a threat that does not exist yet. [34] Anthropic says the big data transfers it needs would make it easier to spot. [33]

Adobe, which makes Photoshop and other design software, published fixes on Tuesday for about 176 flaws across 11 products. [38] Photoshop alone got 18 fixes, nearly all for flaws that could let an attacker run code. [38] Adobe says it has been using advanced AI models to hunt for flaws in its own code. [38] People still check every AI-written fix before it ships, it says. [38]

Why it matters. Adobe says nobody is known to be using any of these flaws. [38] The rise in discoveries has led Adobe to stop publishing its security notices on a monthly schedule. [38]

VMware, whose software lets one computer run many virtual ones, says AI found so many flaws in its products that it now ships updates every month. [46] A VMware manager told The Register that customers struggled to keep up, because each fix takes time to test. [46] Its answer is HyperArmor, a paid add-on. [46] It blocks attacks on a known flaw before the fix is installed, and adds a second login check. [46]

Why it matters. VMware is betting customers will pay extra for it, including for a second login check. [46] Brian Krebs reports that huge monthly fix lists are also straining the teams that test and install Microsoft's updates. [2]

Google said on Tuesday it is raising some of the rewards it pays outsiders for reporting flaws in Chrome, its web browser. [39] In April it cut many rewards, some to a tenth, to steer researchers toward reports it could act on. [39] The base reward for one common kind of memory flaw is now $2,000, four times April's level, but still below what it paid before. [39] The most serious flaws can earn up to $20,000. [39]

Why it matters. AI is changing how many flaws get found and reported, SecurityWeek says. [39] Google recently closed a separate reward scheme for open-source flaws after a rise in AI-written reports. [39]

AIR, an AI security start-up, found a flaw in Vercel Sandbox, a tool companies use to run AI agents in a closed-off space. [41] Agents are AI programs that take actions on their own. [41] The sandbox only lets agents talk to approved servers. [41] AIR showed a specially made security certificate could make an approved server pass data on to an attacker's server. [41]

Why it matters. The sandbox never talks to the attacker's server, so its own checks see nothing wrong, AIR says. [41] Vercel told the researchers it does not yet know how to fix it. [41]

Kimsuky, a North Korean spying group, has been sending fake conference invitations and research requests to Ukrainian experts on security and foreign policy. [42] Genians, a South Korean security firm, found the campaign. [42] In one case they invited a Ukrainian expert to speak in Seoul and offered to pay for travel, with no harmful link at first. [42] Most of the attacks it saw failed. [42]

Why it matters. North Korea has sent troops to fight for Russia in Ukraine since 2024. [42] Genians thinks the spies want to know about North Korean losses and captured soldiers. [42]

XWorm is spying software that gives criminals full control of a Windows computer. [40] Netskope, a security firm, says its makers have added plug-ins for tracking a victim's location, listening through the microphone and watching through the webcam. [40] One plug-in takes over a browser that is already logged in, which skips the second login check. [40] Netskope counted 1,200 incidents at 100 organisations in three months. [40]

Why it matters. Use of XWorm has grown since a cracked copy of its builder was released in June 2025 and its code leaked that August. [40] Netskope expects more upgrades. [40]

Researchers found three flaws in some LG smart TVs that together give full control to anyone on the same home network. [45] No PIN and no action by the owner are needed. [45] The attack starts at the service the LG ThinQ phone app uses to pair with the TV. [45] LG has issued a fix, which was tested on two TV models. [45]

Why it matters. Owners of these TVs are told to check for a firmware update, and there is no workaround for those who cannot update yet. [45] Nobody knows whether other models are affected. [45]

Someone claiming the break-in at Denmark's population register told Politiken, a Danish newspaper, they got in through a company account protected by the password 123456. [43] The register holds the names, addresses and ID numbers of 8.8 million people taken last month. [43] TV 2, a Danish broadcaster, says the attacker ran 14 million automatic searches between 10 and 20 September. [43] Staff spotted it on 2 October from an unusually large bill. [43]

Why it matters. The claim is not confirmed, and Denmark's government did not answer questions about the password. [43] The company whose access was used, Pays ApS, has taken its website offline. [43]

Ledger makes small devices that hold the keys to people's crypto coins. [44] On 13 October it said every confirmed theft so far came from devices sold by one reseller in Southeast Asia. [44] It appears someone along the supply chain soldered a hidden part inside. [44] Ledger says the part reads the 24 secret recovery words off the screen and sends them out over a phone network. [44] An outside investigator says losses may top $86 million, a figure Ledger has not confirmed. [44]

Why it matters. Ledger tells anyone who bought from that reseller in the last 90 days not to set the device up, and to move their coins to a new one. [44]

England's Housing Ombudsman, which settles disputes between social housing tenants and their landlords, is investigating a security breach at an outside supplier. [37] The breach hit its case system on Tuesday, so tenants cannot file new complaints through it. [37] It says the problem is contained and it is checking whether personal data was taken. [37] It has told the UK's data watchdog. [37]

Why it matters. Tenants can still complain by phone or email. [37] The service has told tenants to be wary of unexpected messages asking for personal details. [37]

Only a quarter of security teams use AI agents in their defences, a report by Flashpoint, a threat research company, says. [47] Another 20% are running closed trials, and 35% expect to try them within two years. [47] Those that use them mostly have them look up information and sort alerts. [47] Many will not let an agent take actions on its own. [47]

Why it matters. One respondent said an agent that is confidently wrong is worse than a slow one. [47] Teams also worry about data leaking and being able to check what an agent did. [47]