Microsoft fixes more than 1,000 flaws for a second month, one already used in attacks on Outlook

The stories

26 on this beat today.

Every story told the same way: what happened, who is involved, why it matters, how it works, the truth and the good news.

1. 01

   Microsoft fixes more than 1,000 flaws for a second month, one already used in attacks on Outlook

   1,093

   flaws Microsoft fixed on Tuesday

   the second month in a row above 1,000 [2]

   5,100+

   security fixes from Microsoft so far this year

   a year ago it fixed about 100 a month [1]

   217

   flaws CISA added this year to its list of ones used in attacks

   against about 95,000 flaws given ID numbers this year, Ivanti's Chris Goettl says [1]

   6.5 of 10

   Microsoft's score for the Outlook flaw attackers already use

   it can hand over a user's login [4]

   What happened

   • Microsoft fixed 1,093 security flaws in Windows, Office and its other software on Tuesday. [1]
   • Attackers were already using one of them, in Outlook, Microsoft's email program. [2]
   • It is the second month running above 1,000, after more than 1,400 in October. [2]
   • The Outlook flaw lies in how the program opens a specially made message file. [3]
   • Opening it makes Outlook connect to an attacker's computer without its usual warning. [3]
   • That hands over a hash, a scrambled value made from the user's Windows password. [4]
   • With it, an attacker may sign in as that person elsewhere, or crack the password. [4]
   • The victim must open the file, and Outlook's preview pane does not set it off. [3]
   • Microsoft rated it moderate, 6.5 out of 10. [3]
   • Microsoft credits 2 researchers, and one of them, Benoit Sevens, used Anthropic's Claude AI. [3]
   • CISA, the US cyber-defence agency, added it on Tuesday to its list of flaws used in attacks. [9]
   • Microsoft has now shipped more than 5,100 security fixes this year. [1]
   • A year ago it was fixing about 100 a month. [1]
   • Microsoft says AI tools helped it find and fix so many flaws in recent months. [1]
   • Two other flaws need no click at all, and each is rated 9.8 out of 10. [5]
   • One is in Windows Search, which Exchange email servers use to index incoming attachments. [5]
   • A crafted attachment can run code there, because the server reads it by itself. [5]
   • The other is in LDAP, a Windows network service, and could hit the servers that run logins. [4]
   • Dustin Childs of the Zero Day Initiative, which tracks flaws, counts 30 that could spread on their own. [5]
   • Apple, Google, Mozilla and Adobe also released security fixes this week. [1]
   • Mozilla's Firefox 160 fixed almost 100 flaws on Tuesday. [1]
   • This is the last monthly fix for Windows 11 version 23H2 in its business and school editions. [3]
   • Home and Pro machines on version 24H2 already lost their support last month. [8]

   Who is involved

   • Benoit Sevens a security researcher; Microsoft credits him, using Anthropic's Claude, with finding the Outlook flaw
   • Claude Anthropic's AI model
   • CISA the US civilian cyber-defence agency; it ordered US agencies to fix the Outlook flaw
   • Microsoft the maker of Windows and Office; it fixed 1,093 flaws on Tuesday
   • Dustin Childs head of threat awareness at the Zero Day Initiative, which tracks flaws; he warns that AI speeds up attacks on fresh fixes

   How it unfolded

   1. This month last fix for these editions
   2. From now on no more fixes
   3. Tuesday the fix comes out
   4. End of November US agencies' deadline
   5. Sep Microsoft fixes about 900 flaws [2]
   6. Oct more than 1,400 fixes [2]
   7. 10 Nov 1,093 fixes, one Outlook flaw already in use [2]
   8. End of Nov US agencies must have the Outlook fix in [2]

   Why it matters

   About 95,000 flaws got ID numbers this year, but CISA added only 217 to its list of attacked ones. [1]

   So IT teams must sort over 1,000 fixes to find the few that matter first. [4]

   Childs says AI makes it quick to study a fix and build an attack from it. [5]

   So the danger starts the minute the updates come out, he says. [5]

   How it works

   Security researchers call this patch diffing: comparing old and new code to see what a fix changed. [5]

   The change points straight at the flaw, so an attacker can aim at machines not yet updated. [5]

   Seen it before

   A teacher's red pen on one pupil's test shows the whole class which answers were wrong.

   Why testing finds bugs but cannot show there are none

   Computer scientists call it formal verification: a test checks the cases someone tried, while a proof checks every case.

   The twist

   Microsoft fixing over 1,000 flaws a month tells you how many flaws the testers found. It does not tell you how many are left.

   The picture

   Flaws Microsoft fixed

   A year ago, a month100

   September900

   October1,400

   November1,093

   Step by step

   1. 1 Testers try chosen inputs on the code
   2. 2 A flaw shows up only when a test happens to hit it
   3. 3 Passing tests say nothing about inputs nobody tried
   4. 4 A proof checks a rule against every possible input
   5. 5 But a proof covers only the rules someone wrote down

   The same force, elsewhere today

   Where this chain is also running, in today's other stories.

   • Microsoft's 1,093 fixes

     AI tools tried far more inputs than people could, so they found more flaws, but the count measures the search, not what is still hidden.

   • Chrome's eighth attacked flaw

     Chrome is tested all the time, and attackers still found an input in its XML reader that the testers had not tried.

   • DeepMind's Signal proof

     the proof checked every possible input for the promises it covered, so 2 old bugs put back in made it fail at once.

   • Datadog's cloud count

     the middle number of serious flaws per app rose from 4 to 16 as tools searched harder, so the code was never as clean as the lower count suggested.

   Where you've seen this

   Bridges

   engineers test-load a bridge, but they also work out on paper that it holds every load up to its limit

   School maths

   a rule that works for five numbers you tried is not proved for all numbers

   Train signalling

   some signalling software is proved correct, because no number of test runs covers every train

   The catch

   A proof is only as good as the rules it was asked to check: the Signal proof covered Google's build of the code, not the apps people use.

   And the whole of it

   Every app on a phone was tested by people who could try only so many cases. Very little of it has been proved, and the people who rely on it, us included, cannot see which parts.

   The truth

   • What is really going on Microsoft fixed 1,093 flaws on Tuesday, and attackers were already using one of them, in Outlook. [1]
   • Why it works on people A four-digit count makes the month sound like a flood. Yet CISA has added only 217 flaws to its attacked list this year, against about 95,000 given ID numbers. [1]
   • Childs warns, "It's labeled spoofing, but the impact is bigger than the name suggests." [5]
   • Who gains Attackers who study fresh fixes — Each published fix shows where a flaw was, and Childs says AI makes turning it into an attack near-instant. [5]
   • Who gains Anthropic — Microsoft's advisory credits a researcher who found the Outlook flaw using its Claude model. [3]
   • Who pays IT teams patching Windows and Office — They must work out what actually matters among the fixes, and Knott says teams feel the strain. [4]
   • Who pays Businesses still on Windows 11 version 23H2 — Their business and school editions get no more monthly security fixes after this one. [3]
   • Nobody knows yet

     Who used the Outlook flaw, and against whom.

     Microsoft marked it as exploited but did not say when the attacks started or who was behind them. [5]

   • Nobody knows yet

     How many flaws Microsoft really fixed.

     Outlets count 1,093, 1,094 or 1,235. [1][6] BleepingComputer's 1,235 includes 142 flaws in Edge, Microsoft's browser. [7] October is given as more than 1,400 by The Record and 1,474 by The Register. [2][5]

   Good news

   • Only 1 of the 1,093 flaws Microsoft fixed this month was already being used by attackers, and none had been made public before the fix.

   What comes next

   CISA gave US government agencies until the end of November to install the Outlook fix. [2]

1. 02

   Nitori says 37.6 million records stolen

   What happened

   • Nitori, a big Japanese furniture chain, says an attacker may have taken 37.62 million customer records. [10]
   • The records hold names, emails, home addresses, phone numbers and dates of birth. [10]
   • Account passwords and credit card details were not taken, Nitori says. [10]

   Who is involved

   Nitori runs 879 stores in Japan and 219 more across East and Southeast Asia. [10]

   How it unfolded

   It found the break-in on 6 November, but the attacker had likely been inside since mid-October. [10]

   1. Mid-October the attacker gets in
   2. 6 November Nitori finds it

   Why it matters

   Nitori warns that the stolen details could be used in fake emails and calls to its customers. [10]

   How it works

   Security workers call this phishing: a fake message dressed up to look like it comes from someone trusted.

   The records include past online orders, so a scam email can mention a purchase the customer really made. [10]

   Seen it before

   A stranger who knows your teacher's name and your class sounds like someone from your school.

   The truth

   • Who pays Nitori's customers — Their addresses and order details can make fake emails and calls look genuine. [10]
   • Nobody knows yet

     How the attacker got into Nitori.

     Nitori says it fixed weak points in its systems but has not described the way in. [10]

   What comes next

   Nitori has told Japan's data protection commission and the police, and will contact customers. [10]

2. 03

   Chrome's eighth attacked flaw this year

   What happened

   • Google fixed a flaw in its Chrome web browser that attackers were already using. [11]
   • It is Chrome's 8th attacked flaw this year, the same number as in all of 2025. [12]
   • The bug is in the part of Chrome that reads XML, a common format for data on the web. [11]
   • Someone reported it anonymously on 6 November, and Google released the fix on Monday. [12]
   • Google is keeping the details private until most users have the fix. [11]

   How it unfolded

   1. 6 November flaw reported
   2. Monday fix released

   Why it matters

   Chrome installs the fix by itself, but it only takes effect once the browser restarts. [11]

   How it works

   Programmers call this kind of bug a use-after-free. [11]

   Chrome hands back a piece of memory, then uses it again, which can let an attacker run code. [11]

   Seen it before

   A hotel gives your old room to a new guest, but your old key card still opens the door.

   Chrome is tested all the time, and attackers still found an input in its XML reader that the testers had not tried.

   The idea, in story 01: Why testing finds bugs but cannot show there are none →

   The truth

   • Krebs on Security calls it Chrome's 6th attacked flaw of 2026. [1] The Register counts 8. [12]
3. 04

   Barracuda email boxes attacked after fix

   What happened

   • Attackers are using two flaws in Barracuda's email security gateways, a week after the fixes came out. [13]
   • Used together, the two flaws let an attacker take full control of the device. [13]
   • The attacker first needs to reach the device's admin page, Barracuda says. [13]

   Who is involved

   The gateways are devices that check a company's email for threats before it arrives. [13]

   How it unfolded

   1. Last week Barracuda releases fixes
   2. Monday attacks reported
   3. Tuesday CISA's order
   4. Friday agencies' deadline

   Why it matters

   In 2023, the FBI blamed China-linked attackers for hacking these gateways. [14] 27% of the victims Mandiant found were government bodies. [14]

   How it works

   Security staff call this the patch gap: the days between a fix coming out and it going in.

   Barracuda pushes fixes out automatically, so devices with automatic updates switched off can stay open. [13]

   Seen it before

   A school fixes a broken lock, but classrooms that never got the new lock stay easy to open.

   The truth

   • Barracuda says a small number of organisations were hit, and would not tell The Register how many. [14]
   • Nobody knows yet

     How many Barracuda customers were hit.

     Barracuda says a small number and declined to give The Register a figure. [14]

   What comes next

   CISA gave US agencies until Friday to fix both, and told them to check for break-ins. [2]

4. 05

   Russian spies read government email in secret

   What happened

   • ESET, a security company, says a Russian spy group hid inside government email servers in Kazakhstan and Romania. [16]
   • Its new tool, which ESET calls Cereal, sat inside Microsoft Exchange, the program that runs an organisation's email. [15]
   • It copied chosen emails with document attachments, without stopping normal delivery. [15]
   • In Romania it picked out emails to and from NATO and the European Union. [17]

   Who is involved

   The group, Turla, has been active for more than 20 years and is linked to Russia's FSB security service. [15]

   • Kazakhstan government victim
   • Romania government victim

   How it unfolded

   ESET found it in Kazakhstan on 28 November 2025, and in Romania a month later. [17]

   1. 28 November 2025 found in Kazakhstan
   2. A month later found in Romania

   Why it matters

   Since 2022, about 70% of Turla's target countries were former Soviet states or once close allies, ESET says. [17]

   How it works

   Malware analysts call this living off the land: an intruder using the victim's own programs instead of bringing new ones.

   Cereal had no outside control server; the hacked Exchange server itself took the spies' orders. [16] So defenders had no strange outside address to spot. [16]

   Seen it before

   A spy who reads letters inside the post room leaves no strange van parked outside.

   The truth

   • The link to Russia's FSB rests on years of findings by Western governments and security firms. [16]
   • Nobody knows yet

     How Turla first got into the Exchange servers.

     ESET's research does not describe how the servers were first broken into. [17]

5. 06

   Palantir security chief picked to run CISA

   What happened

   • President Donald Trump has picked Michael LoSapio to run CISA, the main US civilian cyber-defence agency. [19]

   Who is involved

   LoSapio is the chief security officer at Palantir, a data company that sells software to US agencies. [19]

   • Michael LoSapio nominee to run CISA

   How it unfolded

   CISA has had no Senate-approved director in all of Trump's second term. [18]

   It has lost more than 1,000 staff, about a third of its workforce. [20]

   Trump's first pick, Sean Plankey, withdrew in April after more than a year waiting for a vote. [19]

   1. Last year Plankey nominated
   2. April Plankey withdraws
   3. Now a crowded end-of-year Senate
   4. January Democrats take control

   Why it matters

   CISA helps protect power, water and telephone networks, as well as US government computers. [19]

   Democrats due to lead the next Senate and House homeland security committees question Palantir's work for immigration enforcement. [22]

   How it works

   US law limits how long an acting director, a stand-in boss, can serve. [20]

   Nick Andersen's stand-in term ran out on 24 September. [20] Naming a nominee restarts that clock, so he can stay on. [20]

   Seen it before

   A supply teacher can stay longer once the school says a new teacher is on the way.

   The truth

   • A former official questioned whether the government will really push for LoSapio's approval. [20]
   • Who gains Nick Andersen, CISA's acting head — A new nominee restarts the legal clock on his stand-in term, so he can stay in charge. [20]
   • Nobody knows yet

     Whether LoSapio will get a Senate vote this year.

     The US Senate's end-of-year session is crowded, and a nominee not confirmed before it ends must be named again in January. [20]

   What comes next

   The US Senate may not vote before the year ends, and Democrats take control of it in January. [21]

6. 07

   Apple fixes iMessage group chat flaws

   What happened

   • Apple released security fixes for iPhones, iPads, Macs, Apple Watches and Vision Pro headsets this week. [23]
   • One flaw let an attacker join an iMessage group chat and read its messages. [23]
   • Another let a newly added member read messages sent before they joined. [23]
   • A third let someone make a Bluetooth connection to a device with no action by its owner. [23]
   • Apple also sent the fixes to older iPhones still on iOS 18. [1]

   Why it matters

   That covers people whose older devices cannot run Apple's newest system. [1]

   How it works

   Apple calls the group chat bug an authorization issue. [23]

   ZDNet says the app likely let a stranger add themselves without a member's permission. [23]

   Seen it before

   A party door that checks each guest is a real person, but not whether they were invited.

   The truth

   • Apple did not say which kind of Bluetooth attack its fix stops, ZDNet says. [23]
7. 08

   Phone spyware now hits ordinary people

   What happened

   • iVerify, a phone security firm, found 1.07 spyware infections for every 1,000 checks of ordinary people's phones. [24]
   • Work phones showed 0.7 per 1,000, which iVerify puts down to forced updates and company controls. [24]
   • About 3 in 4 infections came from criminals using two leaked attack kits, Coruna and DarkSword. [24]
   • Simply visiting a booby-trapped website could infect a phone with DarkSword, without a click. [24]
   • Coruna was stolen from a defence contractor, whose former executive was jailed for 7 years in February. [24]
   • Apple's spyware alerts reached a bus driver's children and a stay-at-home mum in Ottawa. [25]

   Who is involved

   • A bus driver's children warned by Apple
   • A stay-at-home mum warned by Apple

   Why it matters

   The criminals mostly want logins and crypto, so any phone owner is worth hitting, iVerify's co-founder says. [25]

   How it works

   Security researchers call a ready-made attack kit that many criminals can use commodity malware.

   Once a spy-grade tool leaks into wide use, criminals aim it at crowds instead of chosen targets. [24]

   Seen it before

   A lock pick that only locksmiths owned turns up in a toy shop, and soon every bike shed gets tried.

   The truth

   • iVerify would not say exactly how many phones it checked, only hundreds of thousands. [24]
   • Who pays Ordinary iPhone owners — Criminal spyware now reaches them through websites and adverts, and iVerify found none of those victims got Apple's threat alerts. [25]
8. 09

   SAP fixes four top-rated flaws

   What happened

   • SAP, which makes business software that big companies run on, fixed four flaws rated 10 out of 10. [26]
   • SAP says attackers with no login could use them to take over a system completely. [26]
   • One, which the security firm Onapsis calls ICM2SHELL, sits in the part that receives web traffic. [27]
   • Another, ProxyGate, tricks a gateway into treating an outside attacker as one of its own servers. [26]
   • Onapsis found three of the four flaws. [26] It gave SAP a working attack to prove ICM2SHELL is real. [27]

   How it unfolded

   In August, attackers used another top-rated SAP flaw within days of its fix. [26]

   1. August another flaw fixed
   2. Days later attacks begin

   Why it matters

   A takeover would expose a company's business data and the systems it runs on, Onapsis warns. [26]

   How it works

   Security engineers call this trusting by location: a request is believed because of where it seems to come from.

   ProxyGate works when SAP's list of trusted networks is left open, so outside requests pass as inside ones. [27]

   Seen it before

   A school gate that opens for anyone in the school jumper, pupil or not.

9. 10

   Fireworks AI breach reaches Notion chats

   What happened

   • Fireworks, a company that runs AI models for other firms, disclosed a second October break-in. [28]
   • Attackers hacked a cloud provider hosting its research machines on 23 October, then reached Fireworks the next day. [28]
   • They reached billing details, account records, and parts of some customers' AI models and data. [28]
   • Card numbers were safe, because Fireworks does not store them. [28]

   Who is involved

   Fireworks has more than 10,000 customers. [29]

   How it unfolded

   Fireworks first found a break-in on 6 October. [29] It links that one to a gang that sells stolen AI access. [28]

   1. 6 October first break-in found
   2. 23 October second break-in

   Why it matters

   Notion, a note-taking app, says some of its AI chats may have been exposed, from under 1% of workspaces. [30]

   Notion has stopped using Fireworks and switched off a feature that relied on it. [29]

   How it works

   Security staff call this third-party risk: data is only as safe as each company it passes through.

   Notion's users never chose Fireworks, but their chats passed through its machines. [29]

   Seen it before

   A letter you hand to a friend can still be read if the postman's van is broken into.

   The truth

   • Chief executive Lin Qiao says attackers were not after AI models, because "most downloads were partial and stopped." [30]
   • Who pays Notion users — Their AI chats passed through Fireworks's machines, a company they never chose. [29]
10. 11

    Railway breach ran up $740,000 in AI bills

    What happened

    • Hackers who broke into Railway used stolen keys to run up about $740,000 of AI use. [31]
    • Railway says the attackers were fully removed on 21 October. [31]
    • Engineers and attackers fought for about 26 hours, with the attackers using AI agents. [31]
    • Up to 100,000 workspaces, at most 1.85% of the total, may have had their secrets exposed. [31]

    Who is involved

    Railway is a cloud service where companies host their apps. [31]

    How it unfolded

    The attack began on 20 October, through a flaw in the systems that build customers' apps. [31]

    1. 20 October attackers get in
    2. 21 October attackers removed

    Why it matters

    Railway believes the attackers wanted AI keys to resell or to copy AI models. [31]

    Railway links it to a wave of attacks on AI developer platforms that Wiz, a security firm, calls Black October. [31]

    How it works

    Security engineers call this isolation: each customer's build runs in its own sealed box. [31]

    The attackers broke out of that box, took an internal login, and read Railway's own code for more ways in. [31]

    Seen it before

    A pupil who slips out of the exam room finds the answer sheets in the staff room.

    The truth

    • Railway says 100,000 is the most that could be affected, not the number it confirmed. [31]
    • Who gains Gangs that resell AI access — Railway's attackers went looking for AI keys first, because paid AI access can be resold. [31]

    Good news

    • Railway says it will pay any fraudulent AI charges that the AI companies refuse to refund. [31]
11. 12

    AI helps prove Signal's encryption holds

    What happened

    • Google DeepMind used AI agents to prove with maths that key security promises of the Signal protocol hold. [32]
    • The proven promises include forward secrecy, which keeps old messages safe even if a key is stolen later. [33]
    • A team of 5 researchers did the work in 4 weeks, which Google says used to take specialists years. [33]
    • AI agents wrote about 6.5 million lines of proof, and a checking program rejected any wrong step. [33]
    • The agents flagged about 1% of the code for a human to look at again. [35]
    • As a test, the team put back 2 old, already-fixed Signal bugs, and the proof then failed. [33]

    Who is involved

    The Signal protocol is the encryption used by the Signal app, WhatsApp and Google Messages. [32]

    • Signal
    • WhatsApp
    • Google Messages

    Why it matters

    Signal's chief technology officer, Ehren Kret, says the work gives Signal more confidence in its privacy promises. [32]

    How it works

    Computer scientists call this formal verification: using maths to show that code keeps its promises. [32]

    Each of the 6.5 million lines had to pass the checker, so one wrong step would sink the proof. [33]

    Seen it before

    A few spoonfuls of soup tell you a little; knowing every ingredient that went in tells you all of it.

    the proof checked every possible input for the promises it covered, so 2 old bugs put back in made it fail at once.

    The idea, in story 01: Why testing finds bugs but cannot show there are none →

    The truth

    • The proof covers Google's own build of the code, not Signal's apps, and leaves out some side channels. [33]
    • Google would not say what the work cost. [32]
    • The Register notes that the result assumes the underlying encryption maths is perfect. [34]
    • Who gains People who use Signal, WhatsApp and Google Messages — The core promises of the encryption they rely on are now backed by a checked proof, for the parts it covered. [33]

    Good news

    • The team is releasing its proof and its list of rules so others can build on them. [32]
    • Google DeepMind proved that the encryption used by Signal, WhatsApp and Google Messages keeps its core promises, in four weeks of work by five researchers.
12. 13

    Attackers use a login bypass in Xen Orchestra

    What happened

    • Attackers are using a flaw in Xen Orchestra to log in as administrators without a password. [36]
    • A specially made cookie, a small token a browser sends, tricks it into reusing someone else's login. [36]
    • From there an attacker can take over the whole host as its all-powerful root user, Vates warns. [36]
    • Vates has released fixed versions, 6.9.1 and 6.10.0. [36]

    Who is involved

    Xen Orchestra, made by the French firm Vates, manages virtual machines, many computers running inside one server. [36]

    Why it matters

    One takeover can reach every virtual machine the host runs. [36]

    How it works

    Security engineers call this an authentication bypass: getting past the login check without the right password. [36]

    Xen Orchestra read an empty token as a reason to fetch an existing login, so the attacker borrowed it. [36]

    Seen it before

    A bouncer waves in anyone with an empty wristband, because someone inside already wears a real one.

    The truth

    • What is really going on The same week, attackers broke into Barracuda email gateways through flaws fixed a week earlier, and into Xen Orchestra systems through a login flaw. [13][36]
    • Vates says these attacks leave no trace in the logs, so only network monitoring can spot them. [36]
13. 14

    Ransom message plays on Vanderbilt computers

    What happened

    • Computers at Vanderbilt University in Nashville reportedly played a recorded ransom threat from a gang called BYOD. [37]
    • The message told the university to make contact or see its systems brought down. [37]
    • By Tuesday, Wi-Fi, Outlook, Zoom and some other systems were safe to use again, Vanderbilt said. [38]
    • Classes went ahead, but most systems stayed restricted. [37]
    • Students could not claim tickets for Saturday's football game against Alabama, because ticketing was down. [37]
    • BYOD also claimed an attack on Texas Health and Human Services, which has not confirmed one. [37]

    Who is involved

    • Texas Health and Human Services not confirmed

    How it unfolded

    1. Saturday the game against Alabama
    2. This week ticketing down

    Why it matters

    The gang claims over 1 million US Social Security numbers, the ID numbers Americans use for tax and benefits. [39]

    The truth

    • Vanderbilt has not confirmed the gang's claims about what it took. [37]
    • Who pays Vanderbilt students and staff — They lost the use of most university systems while the attack was cleaned up. [37]
    • Nobody knows yet

      Whether the BYOD gang really holds Vanderbilt's data.

      The claims come from the gang, and Vanderbilt has not confirmed them. [37]

14. 15

    Grok Build share links left data open

    What happened

    • Grok Build, an AI tool that writes apps from plain requests, left thousands of users' data open to anyone. [40]
    • Claroc, a security firm, found 43,601 share links, and more than 35,000 opened without any login. [40]
    • They held medical exam results for 3,784 workers at a construction company in Peru. [40]
    • 24 apps built by SpaceXSI itself were open too, including a confidential plan for a new sales hire. [40]
    • SpaceXSI took down its own 24 links within hours of Ars Technica's questions. [40]

    Who is involved

    It is made by SpaceXSI, the merged SpaceX and xAI, owned by Elon Musk. [40]

    • SpaceXSI maker of Grok Build
    • Elon Musk owner

    How it unfolded

    1. Sunday Claroc warns SpaceXSI
    2. Hours after Ars asks links removed

    Why it matters

    Most other users' data was still open when Ars Technica published. [40]

    How it works

    Ars Technica calls the idea capability-based access: holding the secret link is the permission. [40]

    Grok put each link's random part where public address-lookup records list it. [40] So anyone could collect the links without being sent one. [40]

    Seen it before

    A key under the doormat is only safe until someone publishes a map of every doormat.

    The truth

    • Claroc says no one at SpaceXSI answered its warning on Sunday. [40]
15. 16

    Binance warns of an iPhone app flaw

    What happened

    • Binance, a crypto exchange, warned of an iPhone flaw that lets one app read another app's files. [41]
    • Those files can include wallet recovery phrases, secret keys and logins. [41]
    • It affects iOS versions 26.7.1, 27.0.1 and 27.2, Binance says. [41]
    • Attack code using the trick was published this week, as an app called WQ Files. [41]
    • The app must be on the iPhone for the attack to work, and could be spread outside Apple's App Store. [41]
    • Binance says no crypto losses have been confirmed. [41]

    Why it matters

    The risk covers every iPhone app that stores secrets, not only Binance's. [41]

    How it works

    Phone makers call each app's walled-off space a sandbox, and this trick is a sandbox escape. [41]

    Each app is meant to see only its own files. [41] An escape lets one app reach into the others' storage. [41]

    Seen it before

    Hotel guests each get one room key, so a key that opens every room breaks the whole idea.

    The truth

    • It is not clear whether Apple's 9 November fix closes this flaw, BleepingComputer says. [41]
16. 17

    Open UK government code held 100 serious flaws

    What happened

    • A researcher found more than 100 serious flaws in code that UK government bodies publish openly. [42]
    • He found flaws in about two-thirds of the 349 codebases he scanned, from 33 organisations. [42]
    • The worst: the Planning Inspectorate, which handles land use in England, left a login on GitHub. [42]
    • That login would have opened all its SharePoint sites and staff OneDrive files, he says. [42]

    Who is involved

    Chris Nesbitt-Smith of ControlPlane, a security firm, used consumer AI models on a laptop. [42]

    • Chris Nesbitt-Smith ControlPlane researcher
    • Planning Inspectorate left a login on GitHub

    Why it matters

    Some now argue UK bodies should stop publishing their code. [42] NHS tech teams were reportedly told in May to move away from open code. [42]

    How it works

    Security people call hiding code instead of fixing it security through obscurity.

    Nesbitt-Smith says attackers have likely copied the code already, so closing it now only stops defenders looking. [42]

    Seen it before

    A sheet over a hole in the fence hides it from neighbours, not from the fox that already knows.

    Good news

    • The inspectorate says its tools spotted his activity in August. [42] It fixed the flaw within 20 minutes of the report. [42]
17. 18

    Check Point ties three Chinese spy campaigns

    What happened

    • Check Point, a security firm, says three spying campaigns it tracked separately share one toolkit, called Sparrow. [43]
    • They hit phone companies, governments, militaries and researchers in Latin America, Asia and Europe over the past year. [43]
    • One campaign planted code on real websites that showed visitors a fake Chrome certificate warning. [43]
    • The warning told them to install a Certificate Fix tool, which really installed a backdoor. [43]
    • The hijacked sites served government, shipping and research staff in places including the Czech Republic, Germany, France and Taiwan. [43]

    Who is involved

    It calls the maker CopperDragon, a supplier of tools to several China-linked hacking groups. [43]

    Why it matters

    The targets include phone networks and government offices that whole countries rely on. [43]

    How it works

    Threat analysts call this attribution: matching shared code and servers to link attacks to one source.

    The three campaigns used shared servers and similar code, which is how Check Point tied them together. [43]

    Seen it before

    Three break-ins with the same rare lock-pick marks point to the same toolmaker.

    The truth

    • Check Point says how CopperDragon relates to Salt Typhoon, another China-linked group, is still unclear. [43]
18. 19

    Fake court notices carry spy software

    What happened

    • Blind Eagle, a Spanish-speaking hacking group active since 2018, is sending fake court notices across Latin America. [44]
    • The emails, in Spanish and Portuguese, look like messages from courts, prosecutors and officials. [44]
    • One fake page posed as Colombia's Office of the Attorney General. [44]
    • Opening the attachment can install a remote access tool that lets attackers watch and control the computer. [44]

    How it unfolded

    1. 2018 Blind Eagle first active
    2. Now fake court notices

    Why it matters

    It targets people who handle court cases, contracts and official requests, whose files are sensitive, Zenox says. [44]

    How it works

    Security researchers call this a lure: a message shaped to fit the reader's own job. [44]

    A court clerk expects court papers, so a fake summons gets opened where a random email would not. [44]

    Seen it before

    A note that seems to come from your coach is the one you open first.

    The truth

    • There is no evidence yet that the attacks have succeeded against real targets, The Hacker News reports. [44]
19. 20

    NEAR AI breached through AI server software

    What happened

    • An attacker used a flaw in SGLang, free software that runs AI models, to break into NEAR AI. [45]
    • It happened last Sunday, on one of the servers NEAR AI uses to answer AI requests. [45]
    • The attacker took chat history and settings from 18 AI agents, and names and emails of about 100 users. [45]
    • Two customers' access keys turned up in logs, and NEAR AI cancelled both. [45]

    Who is involved

    NEAR AI sells AI services, including a private mode meant to hide requests even from the company. [45]

    How it unfolded

    1. Last Sunday one server broken into
    2. This week NEAR AI reports it

    Why it matters

    Brave, a web browser maker, runs its Leo assistant's private chats through NEAR AI. [45]

    How it works

    Engineers call the private mode a trusted execution environment: a sealed area of the chip. [45]

    But it guards against outsiders, not against a flaw in the software running inside it, NEAR AI says. [45]

    Seen it before

    A safe keeps thieves out, but not when the thing locked inside is the thief.

    The truth

    • Nobody knows yet

      Whether private chats from Brave's Leo assistant passed through NEAR AI's hacked server.

      NEAR AI has not said whether that server handled any of them. [45]

20. 21

    Backup servers hit through flaws in AhsayCBS

    What happened

    • Huntress, a security firm, says attackers used two AhsayCBS flaws to plant hidden tools on backup servers. [46]
    • The attacker left 5 kinds of hidden web tools, 2 crypto-mining programs and a custom backdoor. [46]
    • On a nearby Veeam backup server, the attacker tried to pull out stored passwords. [46]

    Who is involved

    AhsayCBS runs backup servers, the machines whose job is to hold copies of a firm's data. [46]

    How it unfolded

    Ahsay fixed the flaws in mid-October, but they were used before any fix existed. [46]

    1. Before the fix flaws used
    2. Mid-October Ahsay's fix

    Why it matters

    Backup software holds passwords for many other systems, so the attacker seemed to target it on purpose, Huntress says. [46]

    How it works

    Security staff call a flaw that is attacked before any fix exists a zero-day. [46]

    These flaws were zero-days, so firms that patched in October may already have been entered before the fix. [46]

    Seen it before

    A burglar who got in before the new lock was fitted may have left a window open behind them.

    The truth

    • Huntress cannot say how many of those passwords were taken. [46]
21. 22

    Git fixes a flaw in copying code

    What happened

    • Git, the tool programmers use to copy and track code, fixed a flaw in version 2.56.1. [47]
    • A booby-trapped project could run an attacker's code on a computer that copies it with all its parts. [47]
    • A very long file path passes Git's safety check, then gets saved under a shorter, dangerous name. [47]
    • GitHub, a site where programmers share code, has already fixed its own service. [47]
    • Git for Windows had no fixed version yet on Tuesday evening. [47]

    Who is involved

    • GitHub already fixed

    Why it matters

    Any app that ships its own copy of Git needs updating too, GitHub says. [47]

    How it works

    Programmers call this an integer truncation: a number too big for its box gets cut short. [47]

    Git checked the full long path, then stored a cut-down name that the attacker had chosen. [47]

    Seen it before

    A form box too small for your surname cuts it short, and the short version may be someone else's name.

22. 23

    N-able sees scanning after flaws posted

    What happened

    • N-able says it is seeing scanning linked to the attack code posted online for its N-central systems. [2]
    • N-able took its hosted service offline and told customers to switch off their own servers. [2]
    • It expected a fix around midday on 10 November. [48]
    • It says it has no evidence that any customer was hacked. [2]

    Who is involved

    N-central is a tool IT teams use to manage and control other computers remotely. [2]

    How it unfolded

    A researcher posted more than 30 flaws on GitHub on Sunday. [48] N-able got no advance notice, TechCrunch says. [49]

    1. Sunday flaws posted
    2. 10 November fix due

    Why it matters

    Until the fix is in, companies that use N-central cannot manage their computers through it. [2]

    The truth

    • N-able did not answer TechCrunch on whether it has the logs to check for hacks. [49]
23. 24

    Adobe fixes 110 flaws in 9 products

    What happened

    • Adobe fixed about 110 flaws across 9 products, including Photoshop and Premiere, on Tuesday. [50]
    • Six flaws in Adobe Connect, its online meeting software, were rated 10 out of 10. [50]
    • Adobe says it knows of no attacks using any of them. [50]
    • Krebs says nearly half of Adobe's flaws were rated high severity. [1]

    Why it matters

    Companies running Adobe's Experience Manager websites and Connect meetings had the most serious fixes to install. [50]

    How it works

    Security teams call this a priority rating: a maker's guess at which fixes attackers will reach for first.

    Adobe gave Connect and Experience Manager its middle priority, and rated the rest unlikely to be attacked. [50]

    Seen it before

    A teacher marks which homework mistakes will cost marks in the exam, so those get fixed first.

    Good news

    • Adobe said none of the 110 flaws it fixed on Tuesday had been used in attacks.
24. 25

    Datadog says cloud flaws quadrupled

    What happened

    • Datadog, a cloud monitoring firm, says the typical number of serious flaws in cloud apps has risen fourfold since May. [51]
    • The middle figure went from 4 in May to 14 in June and 16 in November. [51]
    • The count comes from code libraries in apps that Datadog's own customers run. [51]
    • Nearly 30% of AWS and Google Cloud users have access keys older than 3 years, it says. [51]

    How it unfolded

    Datadog ties the jump to May, when Anthropic released its Mythos AI model. [51]

    1. May Mythos released
    2. June middle figure jumps to 14

    Why it matters

    Old keys that never expire are one of the commonest causes of cloud break-ins, Datadog says. [51]

    How it works

    Statisticians use the median: the middle value, so a few extreme cases cannot drag it.

    So in November, half of the apps Datadog watched had 16 or more serious flaws. [51]

    Seen it before

    The middle height in your class says more than the average when one pupil is a basketball star.

    the middle number of serious flaws per app rose from 4 to 16 as tools searched harder, so the code was never as clean as the lower count suggested.

    The idea, in story 01: Why testing finds bugs but cannot show there are none →

    The truth

    • These are Datadog's own figures, drawn only from its customers' apps. [51]
25. 26

    Free AI scans find 475 serious gaps

    What happened

    • Wiz, a cloud security firm, says its free scanning service found more than 475 serious security gaps. [52]
    • It checks charities, hospitals, schools and open-source projects that have no security team of their own. [52]
    • Since September it has watched more than 326,000 public web addresses across 17,700 domains. [52]
    • Finds included exposed admin passwords at national archives and unprotected alert channels at public hospitals. [52]

    Who is involved

    Wiz built it with Google DeepMind and CISA, the US cyber-defence agency. [52]

    • Wiz
    • Google DeepMind
    • CISA

    Why it matters

    Many finds were chains of small gaps, such as a leaked key plus a missing access check. [52]

    How it works

    Analyst Neil Osnato calls it the gap between validation and authority: proof is not permission to act. [52]

    Wiz can prove a hospital's key is exposed, but only the hospital can decide to cancel it. [52]

    Seen it before

    A neighbour can see your window is open, but only you can decide to close it.

    Good news

    • The scans are free for the groups least able to pay for security of their own. [52]