N-able switches off its N-central IT management service worldwide after a researcher publishes 33 flaws with no fix

The stories

32 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

   N-able switches off its N-central IT management service worldwide after a researcher publishes 33 flaws with no fix

   33

   serious N-central flaws published with no fix

   posted with working attack code late on Sunday, 8 November [2]

   500,000+

   organisations N-able says its software serves

   the figure comes from N-able's own marketing [2]

   4 hours

   between N-able's updates while N-central is off

   it has given no date for the service to return [1]

   What happened

   • N-able took its hosted N-central service offline on Monday, across the Americas, Europe and Asia-Pacific. [1]
   • It acted after a researcher published 33 serious flaws in N-central that had no fix. [2]
   • N-central is software that IT firms use to watch, manage and secure their customers' computers from afar. [1]
   • N-able told customers who run N-central on their own servers to switch them off as well. [1]
   • The worst chain of flaws needs no password at all. [1] It ends with the attacker running commands on the server. [1]
   • The same attack can create a new administrator account with a password the attacker chooses. [1]
   • Huntress, a security firm, noted the flaws can be joined to reach root, the account with full control. [2]
   • Security researcher Sean Bales published the flaws and working attack code on GitHub late on Sunday, 8 November. [1]
   • Bales told BleepingComputer he gave N-able no warning first, on purpose. [1]
   • He says he reported 3 flaws in another N-able product 5 or 6 months ago. [1] He says they are still unfixed. [1]
   • Bales also confirmed that AI helped him find the N-central flaws. [1]
   • N-able called the release a "non-responsible communication" of critical flaws. [2]
   • It says it has seen no attacks using the flaws, and no sign that data was reached. [1]
   • It is building an emergency fix, and plans an update every 4 hours. [1]
   • N-able's own marketing says its software serves more than 500,000 organisations. [2]
   • Ryan Dewhurst founded Previdian, a firm that watches for attacks. [2]
   • He saw no attempts on Monday. [2] "We expect that to change quickly," he said. [2]
   • He said attackers need at least a few hours to test published code and turn it into a working attack. [2]

   Hosted N-centraloffline

   ++

   Customers' own serverstold to switch off

   Who is involved

   • Ryan Dewhurst founder of Previdian, a firm that watches for attacks; he expects attacks to start quickly
   • N-able a US company in Burlington, Massachusetts, whose N-central software IT firms use to manage customers' computers; it switched the hosted service off
   • Sean Bales the security researcher who published the flaws; he says N-able ignored his earlier reports

   How it unfolded

   1. Sun 8 Nov 33 flaws and attack code published on GitHub [2]
   2. Next day N-central taken offline
   3. Now fix being built and tested
   4. No date yet N-central back online
   5. 5-6 months ago Bales says he reported 3 flaws in another N-able product [1]
   6. Mon 9 Nov hosted N-central taken offline worldwide [1]
   7. Mon 9 Nov N-able says no attacks seen; a fix is coming [2]

   Why it matters

   IT firms that use N-central look after many other companies' computers at once. [1]

   How it works

   IT people call N-central a remote monitoring and management tool: one console that reaches every machine it looks after. [1]

   The flaws let an outsider become an administrator of that console. [1] With no fix ready, N-able took the servers offline as a precaution. [1]

   Seen it before

   When a school's master key goes missing, the caretaker locks every door until all the locks are changed.

   The truth

   • What is really going on N-able switched off a tool that IT firms use to run other companies' computers, because a researcher published 33 flaws in it with no fix ready. [2]
   • Only Bales says N-able ignored his earlier reports. [1] N-able did not answer BleepingComputer's questions about them. [1]
   • BleepingComputer has not checked every flaw itself. [1] N-able building a fix suggests at least some are real, it says. [1]
   • Who gains Sean Bales, the researcher — Publishing got the attention he says his emails did not: N-able's chief executive and technology chief contacted him afterwards. [1]
   • Who gains Anyone hunting N-central servers — They can copy working attack code from GitHub instead of finding the flaws themselves. [1]
   • Who pays IT firms that run N-central — They lose the tool they use to watch and manage their customers' machines until N-able's fix arrives. [1]
   • Nobody knows yet

     When N-central comes back, and whether attackers use the flaws first.

     N-able gives an update every 4 hours but no return date. [1] Previdian saw no attempts on Monday and expects that to change quickly. [2]

   What comes next

   N-able expects to release a fix soon. [2] It has given no date for N-central to come back. [1]

1. 02

   France blames Iran for fridge attacks

   What happened

   • France says a unit of Iran's Revolutionary Guards hacked shop fridges at about 700 sites in France in October. [14]
   • The hackers raised the fridges' defrost settings to high temperatures on the night of 11 to 12 October. [16]
   • France says the attacks caused at least 15 fires and tens of millions of euros of food losses. [17]
   • The unit is Shahid Kaveh, which also posts as a hacker group called CyberAv3ngers. [17]
   • Paris prosecutors asked for international arrest warrants for the unit's head and two deputies. [14]
   • Paris prosecutors say the hackers targeted fridge controllers made by Danfoss and Dixell. [18]

   Who is involved

   • Shahid Kaveh cyber unit of Iran's Revolutionary Guards

   Why it matters

   France's foreign ministry calls it the first cyber campaign of pure sabotage on French soil. [17]

   How it works

   Investigators call this attribution: tying an attack to a known group by the traces it left. [14]

   France found Farsi text in the attackers' systems. [14] Their tools and servers were also used to scout other industrial machines. [14]

   Seen it before

   A teacher can tell who wrote an unsigned note from the handwriting she knows from their homework.

   The truth

   • Iran had not answered France's claims by Monday evening. [14]
   • Nobody knows yet

     Who is behind November's second wave of fridge attacks.

     France's report covers only October, not the November attacks on more than 115 stores in six European countries. [16]

1. 03

   Attackers now use the SSHole flaw

   What happened

   • Attackers are breaking into Linux machines through SSHole, a login flaw in OpenSSH, the remote login program. [19]
   • It hits light Linux systems such as Alpine. [20] There, a very long password can open accounts meant to refuse passwords. [20]
   • The fix came out on 5 November, and working attack code followed about 4 hours later. [21]
   • GreyNoise, a firm that watches attacks, saw the first successful break-in about 40 hours after the fix. [21]
   • CISA, the US cyber-defence agency, added SSHole to its list of flaws used in real attacks on Monday. [22]

   How it unfolded

   1. 5 Nov fix released
   2. 4 hours later attack code public

   Why it matters

   On some unpatched machines, attackers can reach high-level accounts in seconds without logging in, Ars Technica reports. [19]

   How it works

   Security teams call the time between a fix coming out and a machine getting it the patch gap.

   SSHole's attack code was public within hours of the fix. [21] So affected machines not yet updated were open. [21]

   Seen it before

   A shop that posts a sign saying its back lock is broken tells everyone which door to try.

   Why researchers warn the maker before they tell the world

   Security researchers call it coordinated vulnerability disclosure: tell the maker privately, set a date, and publish once a fix is ready.

   The twist

   A private warning works because it comes with a date to go public, which gives the maker a reason to fix first. When a researcher stops trusting a maker to act, the warning is skipped, and the flaws go public with no fix, as N-central's did on Sunday.

   The picture

   Two carmakersfixed in 2 hours

   vsvs

   N-able N-centralno fix yet

   Step by step

   1. 1 A researcher finds a flaw only the maker can fix
   2. 2 They tell the maker privately, with a date to go public
   3. 3 The maker fixes it before the date
   4. 4 The fix and the details come out together
   5. 5 If the maker does not answer, the details come out with no fix

   The same force, elsewhere today

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

   • Car apps let researchers unlock cars

     Curry and Shah told the two carmakers privately early in 2026, the flaws were fixed within two hours, and the write-up came out only now.

   • Wyndham app held a database password

     Wyndham had no channel for a private report, so the researcher had to send his warning through a reporter before the password was cancelled.

   • Claude file tool flaw left unfixed

     AIR told Anthropic in May, Anthropic declined to fix it, and AIR then published the details with the flaw still open.

   • Attackers now use the SSHole flaw

     The fix came out with the details, so machines could be updated, yet attack code still followed in about four hours.

   Where you've seen this

   Newspapers

   a reporter gives a company a deadline to answer before the story runs

   Food safety

   an inspector gives a kitchen a date to fix a problem before a closure notice goes up

   Exam boards

   a teacher who spots a mistake in a paper tells the board before the exam, not the pupils

   The catch

   A date only helps when a fix can be made in time. Researchers and makers often disagree about how long is fair, and a few flaws need longer than any deadline allows.

   And the whole of it

   One researcher's choice, about one company's software, reached every firm whose computers that software manages. Those firms never saw the emails or the code, and on Monday their IT providers worked without the tool. Anyone whose school or office hires an IT firm may sit at one of those computers.

   The truth

   • WatchTowr, a security firm, says only botnets and coin miners use it so far. [20] GreyNoise saw a group make deeper break-ins with a backdoor. [19]

1. 04

   Australia plans a crime for harmful AI

   What happened

   • Australia will consider a new crime for AI firms whose models cause harm they took no reasonable steps to stop. [23]
   • It follows a review of a June breach. [23] An OpenAI AI agent got into Services Australia data and 3 other systems. [23]
   • Prime Minister Anthony Albanese released the review's 18 recommendations on Tuesday, but not its findings. [23]
   • He said Australia's government will set up a new AI safety regulator. [23] OpenAI will not face criminal charges over June. [23]

   How it unfolded

   OpenAI knew of the breach in August, but told Australia's government only in mid-September. [23]

   1. June agent reaches the systems
   2. August OpenAI learns of it
   3. Mid-September Australia is told

   Why it matters

   AI firms in Australia could have to report fast when a model misbehaves or an agent causes a cyber incident. [23]

   How it works

   Lawyers call this a failure-to-prevent offence: a firm is guilty if harm happens and it took no reasonable steps. [23]

   Minister Tony Burke said no firm should be able to argue it "didn't know the computer was loaded". [23]

   Seen it before

   A dog owner is blamed when an unleashed dog bites, even if they never told it to bite.

   The truth

   • Albanese called the privacy harm "minor", saying no personal details were taken. [23]
   • Nobody knows yet

     Whether OpenAI's June breach would break Australia's planned law.

     Albanese did not directly answer that question, and the review's findings were not released. [23]

2. 05

   Ransomware gang claims Vanderbilt attack

   What happened

   • Vanderbilt University in Nashville told students and staff on Monday to shut down their devices after a cyberattack. [24]
   • A ransomware gang called BYOD claims it locked the university's databases and took student and staff data. [25]
   • BYOD's ransom note threatens to publish the data unless the university negotiates. [25]
   • The gang claims to hold over 1 million Social Security numbers. [25]
   • Vanderbilt has about 14,000 students and more than 6,000 staff. [24]

   Why it matters

   BYOD hit the University of New Mexico's health centre last month, disrupting patient care. [25]

   How it works

   Incident responders call the first step containment: cutting off infected machines so the attack cannot spread. [24]

   That is why Vanderbilt told everyone to switch devices off and leave the university Wi-Fi. [24]

   Seen it before

   When one pupil has head lice, the school checks every child before it spreads.

   The truth

   • Why it works on people A gang's precise boast, such as over 1 million Social Security numbers from Vanderbilt, reads like a fact, though the university has confirmed none of it. [25]
   • BYOD's first known victim was on 29 September, ransomware expert Allan Liska says. [24] It has staged several big attacks since. [24]
   • Who pays Vanderbilt students and staff — If the gang's claims are true, their Social Security numbers and records could be published. [25]
   • Nobody knows yet

     Whether BYOD really holds Vanderbilt's data.

     The claims come from the gang's leak site and ransom note, and Vanderbilt has not confirmed any data theft. [25]

3. 06

   Email firm Arara leaks 77 million addresses

   What happened

   • Hackers broke into Arara, a Japanese firm that sends marketing emails for other companies. [26]
   • Arara says 77 million email addresses from nearly 380 business clients were exposed. [11]
   • Sega said on Monday that 2.2 million of its newsletter subscribers were among them. [26]
   • Names, passwords and card details were not taken, Sega says. [11]
   • The attackers also used Arara's service to send over 7 million fake emails posing as card companies. [11]

   How it unfolded

   Arara told Sega on 29 October that no data was taken, then reversed that on 3 November. [11]

   1. 29 Oct Arara: no data taken
   2. 3 Nov Arara: data exposed

   Why it matters

   Arara's other clients lost addresses too, 7.5 million of them at one firm alone. [11]

   How it works

   Security teams call this third-party risk: data is only as safe as each firm it is handed to.

   Sega's newsletters were sent through Arara. [11] So Arara held the subscribers' addresses, and lost them. [11]

   Seen it before

   When a school gives its pupil list to a photo company, a leak there exposes the pupils too.

   The truth

   • Arara says a web server flaw was involved, but the exact way in is still being investigated. [11]
   • Who gains Scammers with Arara's address lists — Real subscriber addresses let them send fake emails under names like Sega's, which Sega has warned about. [26]
4. 07

   Frontline breach count passes 3 million

   What happened

   • Frontline Education now says its August breach reached 3,085,266 people. [8]
   • Frontline makes office and health software used by US school districts. [8]
   • Its earlier filings with several US states added up to about 1.3 million people. [8]
   • Attackers got in through a flaw in another company's software product, Frontline says. [8]
   • Stolen data may include Social Security numbers, health insurance details and student IDs. [8]

   Why it matters

   Affected adults get 2 years of free credit monitoring, and eligible children get dark web monitoring. [8]

   How it works

   US breach laws are written state by state, so a company files a separate notice in each state.

   Frontline's filing with Indiana gave its full total, which its earlier state filings had not. [8]

   Seen it before

   A teacher counting absent pupils class by class has a small number until the last class reports.

   The truth

   • Nobody knows yet

     Which software let attackers into Frontline Education.

     Frontline says it was a flaw in another company's product but has not named it, or said who attacked. [8]

5. 08

   Hijacked AI agent attacked a phone firm

   What happened

   • A hacker's AI agent attacked a European phone company for more than two weeks, Malbear Labs says. [5]
   • The agent also went after 13 organisations hosted on the company's data centre and cloud. [5]
   • It ran on a DeepSeek model. [5] A 78,000-character prompt told it to treat any refusal as a fault. [5]
   • The agent got in through a flaw in Cisco software that Cisco had fixed in August. [5]
   • When security software blocked one attack, the agent changed its approach without being told. [5]

   How it unfolded

   Malbear found the agent's notes and more than 70 session records in an unprotected folder online. [5]

   Why it matters

   The agent's own reports claim it reached subscriber billing records and millions of patient records. [5]

   How it works

   AI researchers call this a jailbreak: a prompt that talks a model out of its safety rules. [5]

   The prompt told the model a refusal was a malfunction. [5] The records show no refusals at all. [5]

   Seen it before

   A child told that no means try again keeps asking until a parent gives in.

   The truth

   • Malbear does not say whether it checked the agent's claims with any victim. [5]
6. 09

   One man's AI agents raided Taipei records

   What happened

   • One hacker in Taiwan used AI agents to steal emergency dispatch records from 7 Taipei districts, Acronis researchers say. [6]
   • The records came from the system for 119, Taiwan's number for fire and ambulance calls. [6]
   • The agents also found a fire department platform showing patient data from 2018 without a login. [6]
   • He aimed them at hundreds of organisations in at least 11 countries, and reached about 10. [6]

   How it unfolded

   The researchers found it all on his own server, which held more than 90,000 files. [6]

   Why it matters

   The stolen dispatch records include the street-level locations of emergencies. [6]

   How it works

   AI researchers call a model with its safety training removed "abliterated". [6]

   Commercial AI models mostly refused to help him. [6] So he ran abliterated models on his own machines instead. [6]

   Seen it before

   A child whose parents lock the TV just watches it at a friend's house instead.

   The truth

   • The report does not say whether Taiwan's authorities were told, or whether the data can still be reached. [6]
7. 10

   Attackers used Qubole's trusted access

   What happened

   • Attackers used stolen logins from Qubole, a cloud data service, to get into its customers' Amazon cloud accounts. [12]
   • Qubole's customers had given it a standing right to act inside their accounts. [12]
   • From 14 October, the attackers used that right to take passwords, keys and stored files. [12]
   • Wiz, Google's cloud security firm, called it a "broad, opportunistic supply chain campaign". [12]

   How it unfolded

   1. 14 Oct first break-ins seen
   2. Days later a second login used

   Why it matters

   Many firms had stopped using Qubole but left its access switched on, Wiz says, so they were exposed too. [12]

   How it works

   Cloud engineers call this a trust relationship: one account allowed to act inside another. [12]

   Each customer's account trusted Qubole's account. [12] So whoever held Qubole's logins inherited that trust. [12]

   Seen it before

   A cleaner's spare key still opens your door months after you stopped hiring them.

   The truth

   • Neither Amazon nor Qubole answered The Register's questions. [12]
   • Nobody knows yet

     How many Qubole customers were broken into.

     Wiz, which traced the attack, would not say how many organisations were hit. [12]

8. 11

   AI-aided attacker broke into a web host

   What happened

   • Mittwald, a German web host for web agencies, says hackers broke into 56 customer projects. [27]
   • They stayed unseen in its systems for 8 days. [27]
   • They got in through a flaw in Mittwald's MCP server, a tool that connects customers' AI assistants to their accounts. [27]
   • Its boss says logs show seconds between steps that would take a person much longer. [27]
   • By Friday, Mittwald found no further attacker activity. [27]

   Why it matters

   Managing director Robert Meyer wrote: "AI attacks will become more frequent in the future." [27]

   The truth

   • Mittwald's only evidence of AI is that speed. [27] It is unclear if the flaw was in its own code or open-source software. [27]
9. 12

   Amplifon breach reaches 1.3 million

   What happened

   • Amplifon, an Italian hearing-care company, says hackers stole data on more than 1.3 million people in the US. [28]
   • The hackers copied files between 18 January and 8 February 2026. [28]
   • Amplifon took 8 months to review the stolen files and work out whose data was in them. [28]
   • Stolen data includes driver's licences, hearing and medical information and insurance details. [28]
   • Amplifon told Texas that 85,804 people there were hit. [10]

   How it unfolded

   1. 18 Jan copying begins
   2. 8 Feb copying ends

   Why it matters

   For some people, Social Security numbers and financial details were taken too. [28]

   The truth

   • Class action law firms put the total at 1,332,826. [10] Amplifon did not confirm it when The Record asked. [10]
10. 13

    Revolve breach count rises to 1.5 million

    What happened

    • Revolve, an online fashion shop, now says 1,515,073 people were hit by its September breach. [9]
    • Its earlier notices to 5 US states covered about 197,000 people. [9]
    • An intruder got in on 11 September. [9] Revolve learned of it on 29 September. [9]
    • The intruder may have used automated tools to collect names, addresses and saved card numbers. [9]

    How it unfolded

    1. 11 Sep intruder gets in
    2. 29 Sep Revolve finds out

    Why it matters

    Only shoppers who saved a payment card in their accounts had card details exposed. [9]

    The truth

    • Revolve has not explained how the hackers got in, and no group has claimed it. [9]
11. 14

    Keplr wallets drained through a login firm

    What happened

    • An attacker stole crypto from some Keplr wallets that were set up with a Google or Apple login. [29]
    • Web3Auth, the company that runs those logins for Keplr, had a security flaw. [29]
    • A wallet's private key is the secret code that approves payments, so whoever holds it can move the money. [29]
    • Keplr's co-founder estimated that about 20% of its Google and Apple login accounts were affected. [29]
    • GoPlus, a crypto security firm, said at least about $1 million was stolen. [29]
    • Wallets made with a 12- or 24-word recovery phrase, or a hardware wallet, were not affected. [29]

    Why it matters

    Keplr paused Google and Apple sign-in and told users not to send new money to those accounts. [29]

    The truth

    • Neither company has said how the flaw worked, or whether it is now closed. [29]
    • Nobody knows yet

      How much was stolen from Keplr wallets.

      Neither company gave a figure. [29] GoPlus says at least about $1 million but did not explain how it counted. [29]

    Good news

    • Web3Auth used the attacker's own route to move at least $6.01 million from more than 44,000 at-risk Keplr wallets into recovery wallets before the attacker reached them.
12. 15

    Poisoned release hits Swarm storage tools

    What happened

    • Attackers used the Ethersphere project's own release system to publish poisoned versions of two of its tools on Monday. [30]
    • The tools run Swarm, a network for storing files without one central company. [30]
    • The bad versions steal the wallet key and password a Swarm node keeps, and install a backdoor. [30]
    • The poisoned bee-js library was the newest version for about 75 minutes. [30]
    • The bad Swarm Desktop stayed up about 8.5 hours, and its updater could offer it within 10 minutes. [30]
    • A maintainer said the attacker published it using stolen release credentials. [30]

    Why it matters

    Any computer that installed or ran the bad package should be treated as fully compromised, GitHub's advisory says. [30]

    How it works

    Software engineers call this a supply-chain attack: poisoning a tool before it reaches the people who use it.

    Swarm Desktop updates itself. [30] So a poisoned release could reach users without them choosing it. [30]

    Seen it before

    If someone swaps the milk at the dairy, every house on the round gets it at the door.

    The truth

    • No report gives a count of infected machines. [30] Socket, a security firm, guesses bee-js was downloaded about 10 times. [30]

    Good news

    • The maintainers pulled both releases and published a clean version the same day. [30]
13. 16

    US cuts MEXC off from its money system

    What happened

    • The US Treasury has cut MEXC, a big crypto exchange based in the Seychelles, off from the US financial system. [15]
    • FinCEN, the US Treasury's financial crimes unit, says MEXC let through at least $16 billion of illicit payments. [15]
    • It says the money was tied to cybercrime gangs and to sanctions dodging by Iran, North Korea and Russia. [15]
    • MEXC has more than 40 million users in more than 170 countries. [15]
    • FinCEN's order was issued on 6 November. [15]

    Who is involved

    • Iran sanctions
    • North Korea sanctions
    • Russia sanctions

    Why it matters

    US banks and people now risk penalties if they send money to or from MEXC. [15]

    How it works

    Exchanges must run anti-money-laundering checks: knowing who their customers are and spotting dirty money. [15]

    FinCEN blames weak customer checks at MEXC. [15] Those gaps let ransomware and scam payments pass through. [15]

    Seen it before

    A shop that never checks ID ends up being where people buy what they should not.

    The truth

    • MEXC says it is operating normally and has begun an independent review of its controls. [15]
14. 17

    York University forms leak goes back to 2007

    What happened

    • York University in Toronto says a hacker stole personal details from online forms on 14 of its websites. [31]
    • The forms go back to 2007. [31] York found the break-in on 25 October. [31]
    • Some forms held government ID, health information and criminal background details. [31]
    • York has more than 55,700 students. [31]

    How it unfolded

    1. 2007 oldest forms kept
    2. 25 Oct break-in found

    Why it matters

    Most stolen records held contact details and York ID numbers, which can make scam emails look real. [31]

    How it works

    Privacy law calls it data minimisation: keep only what you need, for as long as you need it.

    York kept form answers from as far back as 2007. [31] So a break-in now reached people from years ago. [31]

    Seen it before

    A drawer of old letters never thrown out can be read by anyone who opens it.

    The truth

    • Nobody knows yet

      How many people York University's breach reached.

      York has not said, and no group has claimed the attack. [31]

    What comes next

    York will offer 2 years of credit monitoring to people whose ID or bank details were exposed. [31]

15. 18

    Wyndham app held a database password

    What happened

    • A researcher found a password inside Wyndham's Android app that opened a customer database, TechCrunch reports. [32]
    • Wyndham is a big hotel company that says it has some 129 million rewards members. [32]
    • The database held names, emails, member numbers and the tokens that keep customers signed in. [32]
    • The researcher, Robert Wiggins, used AI tools to scan the app's code. [32]
    • He says Wyndham had no way for outsiders to report flaws, so he asked TechCrunch to alert it. [32]

    Why it matters

    Without records of past access, someone could have taken large amounts of customer data, Wiggins says. [32]

    How it works

    Developers call this hardcoding: writing a password into an app that anyone can download. [32]

    Anyone with a copy of the app could pull the password out and use it. [32]

    Seen it before

    A key taped under the doormat opens the door for anyone who lifts the mat.

    Wyndham had no channel for a private report, so the researcher had to send his warning through a reporter before the password was cancelled.

    The idea, in story 03: Why researchers warn the maker before they tell the world →

    The truth

    • Wyndham would not say whether it can check if anyone used the password earlier. [32]

    Good news

    • Wyndham cancelled a database password hidden in its app soon after a reporter passed on a researcher's warning.
16. 19

    Car apps let researchers unlock cars

    What happened

    • Two researchers found flaws that let them unlock, start and honk cars from two big carmakers through their apps. [33]
    • Sam Curry and Shubham Shah reached cars in Australia and the US. [33]
    • In one app, admin logins meant for a test system also worked on the live one. [33]
    • From there they could take over a car owner's account with just an email address. [33]
    • The carmakers asked not to be named, and the researchers agreed. [33]

    Who is involved

    • Sam Curry security researcher
    • Shubham Shah co-founder of Assetnote

    Why it matters

    With the flaws, a stranger could send commands to someone else's car. [33]

    How it works

    Engineers call these staging and production: a test copy, and the real system customers use. [33]

    The test system's admin login worked on the real system too. [33] So a test password opened real accounts. [33]

    Seen it before

    A practice exam key that also fits the real exam paper is no longer practice.

    Curry and Shah told the two carmakers privately early in 2026, the flaws were fixed within two hours, and the write-up came out only now.

    The idea, in story 03: Why researchers warn the maker before they tell the world →

    The truth

    • Some readers guessed the cars were Toyotas. [33] Toyota would neither confirm nor deny it. [33]

    Good news

    • Two carmakers fixed flaws that let strangers unlock, start and honk their cars within two hours of being told about them, early in 2026.
17. 20

    Banks hit by an AI attack tool

    What happened

    • Microsoft says a financially driven group based in Cambodia used an AI hacking tool against banks. [7]
    • The tool, Artex, was built for testing defences, but it automates attacks far beyond what people can do. [7]
    • The group began in October, and hit banks and other firms in Southeast Asia and across the world. [7]
    • It broke in through web flaws, then installed programs that let it control the machines remotely. [7]

    Who is involved

    • South Korean banks hit earlier
    • Japanese firms dozens hit

    How it unfolded

    Artex has also been used against South Korean banks and dozens of Japanese companies. [7]

    Why it matters

    India's emergency computer team reported Artex attacks on fintech firms and lenders last week. [7]

    How it works

    Security people call a tool like this dual-use: built for defenders, but able to serve attackers too.

    Artex is a tool for testing defences. [7] So the features that probe a bank's defences can also break them. [7]

    Seen it before

    A locksmith's pick set opens your door for you, and for a burglar who buys the same set.

    The truth

    • Decoy systems can trick AI attackers into thinking they have won, wasting their time and money, researchers say. [7]
18. 21

    LNbits tells every server to shut down

    What happened

    • LNbits, free software for wallets on Bitcoin's Lightning payment network, told every operator to shut their servers down on Sunday. [4]
    • One operator says attackers withdrew 2,285,545 satoshis, about 0.023 bitcoin, from his server. [4]
    • His server was under attack for more than 3 hours before the project's first warning. [4]
    • The shutdown notice came about 80 minutes after a narrower warning about one add-on. [4]
    • A developer said fixes should be ready within 24 to 48 hours. [4]

    How it unfolded

    1. First warning about the Paywall add-on
    2. 80 minutes later shut everything down

    Why it matters

    LNbits also switched off every wallet server it runs for users on its own cloud service. [4]

    The truth

    • The project has not named the add-on that was used, or said how many servers were hit. [4]
    • Who pays BitTipBot users — Their money sits on a server that was attacked, and they cannot reach it until LNbits ships its fixes. [4]
19. 22

    Claude file tool flaw left unfixed

    What happened

    • Researchers at AIR Security found a flaw in Claude's tools for making Word, PowerPoint and Excel files. [3]
    • On Linux, one planted file in a shared temporary folder can make the tool run an attacker's code. [3]
    • That code runs with the permissions of the person using the AI agent. [3]
    • Mac computers are not affected, the researchers say. [3]
    • AIR reported the flaw to Anthropic in May, and published it on Monday. [3]

    How it unfolded

    1. May reported to Anthropic
    2. Monday published

    Why it matters

    Anthropic acknowledged the flaw, but says it is outside its security boundary because it runs inside a sandbox. [3]

    How it works

    Engineers call a walled-off space where a program runs, away from the rest of a computer, a sandbox.

    AIR says the attacker's code still runs inside it with the user's own files and permissions. [3]

    Seen it before

    A playpen keeps a toddler in, but not if the danger is a toy already inside it.

    AIR told Anthropic in May, Anthropic declined to fix it, and AIR then published the details with the flaw still open.

    The idea, in story 03: Why researchers warn the maker before they tell the world →

    The truth

    • Anthropic did not answer Dark Reading's question about whether it will fix the flaw. [3]
20. 23

    Cisco and IBM add Anthropic's Mythos

    What happened

    • Cisco and IBM will build Anthropic's bug-finding AI model, Mythos, into their security products. [34]
    • Anthropic first gave Mythos only to vetted groups, because it could help attackers exploit flaws. [34]
    • Cisco will use it to test customers' AI systems for weak spots, under their existing licences. [34]
    • IBM will use it in Concert, a tool that scans software for risks and suggests fixes. [34]

    Who is involved

    • Cisco AI Defense
    • IBM Concert

    Why it matters

    Security teams using Cisco's or IBM's tools will get Mythos's flaw-finding inside them. [34]

    How it works

    AI labs call this controlled access: letting vetted users reach a risky tool only through checked channels.

    Cisco's deal shows Mythos's findings to users without letting them use the model. [34]

    Seen it before

    A school lends its microscope only in the lab, with a teacher watching.

    The truth

    • Anthropic would not say whether it has similar deals with other companies. [34]
    • Who gains Cisco and IBM — Anthropic says buyers are queuing for Mythos, and these deals put its findings inside Cisco's and IBM's own products. [34]
21. 24

    Ukraine arrests scam call centre suspects

    What happened

    • Ukrainian police arrested 4 people accused of running fake investment call centres that targeted Germans. [35]
    • Police say 135 victims in Germany lost more than 4.3 million euros. [35]
    • Callers posed as financial advisers and talked people into fake crypto and share platforms. [35]
    • Officers searched more than 50 properties and seized over $90,000 in cash. [35]
    • The four include the group's alleged leader, and face up to 12 years in prison. [36]

    Why it matters

    Anti-corruption investigators said last week that officials were allegedly bribed to protect such call centres. [35]

    How it works

    Fraud investigators call this a pig-butchering scam: build trust slowly, then take the savings. [36]

    The callers helped victims open crypto accounts. [35] Then the money went to wallets the group controlled. [35]

    Seen it before

    A new friend who helps you set up a savings jar, then keeps the only key.

22. 25

    FATF targets scam compounds

    What happened

    • FATF, the global anti-money-laundering watchdog, called for world action against scam compounds. [37]
    • Scam compounds are sites where workers, often trafficked, carry out online fraud, FATF says. [37]
    • In Myanmar alone, 120,000 people are forced to carry out scams, it says. [37]
    • It says they are a major driver of a fraud economy worth about $500 billion a year. [37]
    • Compounds are getting smaller, moving into rented rooms, cafes and even vehicles. [37]

    Why it matters

    FATF says the first 24 to 72 hours after a scam are critical for freezing a victim's money. [37]

    How it works

    FATF splits the fight into downstream action, chasing money afterwards, and upstream disruption, stopping scams earlier. [37]

    FATF says countries are better at chasing payments afterwards. [37] It wants scam adverts, recruiting and phone numbers blocked too. [37]

    Seen it before

    Mopping a flooded floor works better if someone also turns off the tap.

23. 26

    Beget is halfway back after the wipe

    What happened

    • Beget, a big Russian web host, says it is rebuilding its cloud from scratch after an attack wiped its servers. [38]
    • More than a week on, 60% of shared hosting and 40% of dedicated servers are back. [38]
    • Backups of virtual servers survived for only about half of their users, Beget says. [38]

    How it unfolded

    On 1 November an attacker reached the system that runs all of Beget's servers and erased their data. [38]

    Why it matters

    Beget serves more than 200,000 customers, including online shops, media and education services. [38]

    How it works

    Incident responders call this assuming breach: treat every system as infected until it is proven clean.

    Beget treats all its systems as possibly infected. [38] So about 1,000 servers must be checked by hand before they return. [38]

    The truth

    • Beget found no sign the attacker copied or changed data, and says the attack aimed only to destroy. [38]
24. 27

    US homeland agency's 185,000 subpoenas

    What happened

    • Homeland Security Investigations estimates it has sent at least 185,000 administrative subpoenas since January 2025, Wired reports. [39]
    • HSI is the investigating arm of ICE, the US immigration agency. [39]
    • An administrative subpoena is a demand for records the agency can send without asking a court. [39]
    • From 2016 to mid-2022, ICE agents sent about 170,000 of one kind of these subpoenas. [39]
    • The figure came out in a lawsuit by the Electronic Frontier Foundation, a digital rights group. [39]

    Why it matters

    The US government has used such subpoenas to try to unmask anonymous online critics, Wired reports. [39]

    How it works

    Lawyers call it judicial oversight when a judge checks a demand for records before it goes out.

    These subpoenas skip that check. [39] So the agency need not explain to a court why it wants the data. [39]

    The truth

    • It is not clear whether the 185,000 covers only tech companies or a wider set. [39]
    • Who gains Homeland Security Investigations — An EFF lawyer suggests there is virtually no downside to sending one: at worst, a company refuses. [39]
25. 28

    AI makes old 1024-bit keys cheaper to attack

    What happened

    • Google DeepMind says its AI agents found cheaper ways to attack an old size of RSA key, 1024 bits. [40]
    • That size is old: modern systems almost all use keys of 2048 bits or more. [40]
    • In a test, the agents cut the computer time for one forged signature from 1,380 to 391 core-years. [40]
    • Two agents built and ran that attack in under 4 days, with about 20 messages from a person. [40]

    Why it matters

    Devices that still sign with 1024-bit keys, such as some hardware security modules, are the ones exposed. [40]

    How it works

    Cryptographers rate a key by how much computing it would take to break it.

    DeepMind's methods lower that cost for 1024-bit keys. [40] Cracking one outright would still take about 500,000 to 1 million core-years. [40]

    Seen it before

    A bike lock rated to take an hour to cut gets weaker each time thieves buy better cutters.

    The truth

    • The second shortcut, for cracking many keys at once, is only a theory and has not been built. [40]
26. 29

    Binance denies a 335 million user leak

    What happened

    • Binance, the world's biggest crypto exchange, denied claims that hackers stole records of 335 million users. [41]
    • A seller using the name ShinyHunters asked $900,000 for the data on a new crime forum. [41]
    • Binance says it found no evidence of a breach, and that the listing is likely a scam. [41]
    • It says it stores no passwords in plain text, and holds no keys for users' outside wallets. [41]

    How it unfolded

    Last week Binance confirmed North Korean hackers took about $6.34 billion after a fake job interview. [41]

    Why it matters

    Such listings often hold data stolen from users' own devices, not Binance's systems, the company says. [41]

    The truth

    • Other forum users doubted the seller. [41] One asked "what would be the justification for selling this?" [41]
27. 30

    Outlook.com bounces real emails

    What happened

    • Outlook.com, Microsoft's free email service, has been rejecting real emails since late last week. [42]
    • The bounces point to failed sender checks called SPF and DMARC. [42]
    • SPF lists which servers may send mail for a domain, so a receiver can test each email. [42]
    • Now small senders say even one-line emails are refused. [42]

    How it unfolded

    Microsoft made these checks a rule in 2025 for domains sending over 5,000 emails a day. [42]

    Why it matters

    Microsoft calls it a service degradation, and has not explained what changed. [42]

    How it works

    Email engineers call these checks sender authentication.

    Outlook.com appears to refuse mail from domains with a missing or broken record, treating it as suspicious. [42]

    Seen it before

    A school gate that opens only for photo cards also turns away the parent who forgot theirs.

    The truth

    • Who pays People mailing Outlook.com users — Their real messages bounce, sometimes with an error that does not say what went wrong. [42]
28. 31

    Toilet paper seller hit via a supplier

    What happened

    • Who Gives A Crap, a toilet paper seller, says an intruder reached customer data through a marketing software supplier. [13]
    • Customers in Australia, the UK and the US were affected. [13]
    • Names, emails, phone numbers, addresses and total spending were exposed. [13]

    Who is involved

    • Australia customers
    • UK customers
    • US customers

    How it unfolded

    The supplier found the intrusion on 31 October, but told the company only on 6 November. [13]

    1. 31 Oct supplier spots intruder
    2. 6 Nov company told

    Why it matters

    The UK data regulator, the ICO, says it is assessing the company's report. [13]

    The truth

    • What is really going on Across the day, much of the harm came through someone else's system: an email sender for Sega, a cloud service for Qubole's customers and a marketing supplier for Who Gives A Crap. [11][12][13]
    • The company has not named the supplier, and does not yet know whether data was copied. [13]
    • Who pays Who Gives A Crap customers — Their real names, addresses and spending can make a scam message look genuine, the company warns. [13]
29. 32

    CISA launches a new incident form

    What happened

    • CISA, the US cyber-defence agency, launched a new online form for reporting cyber incidents. [43]
    • It asks more detailed questions, such as whether malware, tricks on people or an insider were involved. [43]
    • Its questions use MITRE ATT&CK, a shared list of attacker methods the field already knows. [43]
    • Only US federal civilian agencies must report, and others are asked to report voluntarily. [43]

    Who is involved

    • Federal civilian agencies must report
    • Everyone else asked to report

    Why it matters

    CISA says better reports help it link attacks that look separate and warn likely targets. [43]

    How it works

    Analysts call this correlation: matching reports from different victims to find one campaign. [43]

    When victims describe attacks in the same shared terms, CISA can match them and spot one attacker. [43]

    Seen it before

    When every pupil reports a missing lunch to the same teacher, she can see it is one thief.