Hackers linked to Russia's spy service poisoned websites to reach phones that had missed updates

The stories

27 stories 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

   Hackers linked to Russia's spy service poisoned websites to reach phones that had missed updates

   8

   known flaws the attackers used, all already fixed by Apple or Google

   they cover older iPhones and Chrome on Android [3]

   3

   routes to reach a phone: real sites, fake copies and a tampered tool

   each route delivered the same hidden code [2]

   18 November

   the date US federal agencies are told to patch one of the flaws by

   Apple fixed that flaw in May 2024 [5]

   What happened

   • Hackers hid code on real and fake websites to break into iPhones and Android phones that visited. [1] The attack needed no tap from the visitor. [2]
   • Sekoia, a French security firm, found the campaign in mid-October. [4] It published the work on 4 November. [3]
   • Sekoia ties it, with moderate to high confidence, to a hacking group called APT29. [3] APT29 is widely linked to Russia's Foreign Intelligence Service, its spy agency abroad. [3]
   • The attacks worked only on phones that had missed updates. [2] Apple and Google had already fixed all the flaws. [1]
   • The attackers used 8 known flaws across iPhones and Android phones. [3]
   • The attackers used 3 routes to reach visitors. [3]
   • First, they put code into real websites, including a Ukrainian gas producer and a US chemical trade group. [1]
   • Second, they built fake copies that passed along the real pages with code added. [2] One copied America250, the group marking 250 years of US independence. [2]
   • Another fake copied Raffles London at The OWO, a luxury hotel near Britain's defence ministry. [2]
   • Third, they tampered with Plerdy, a Ukrainian tool that tracks website visits, from 27 August to 3 September. [3] Thousands of mostly Ukrainian websites load Plerdy. [3]
   • That week exposed hundreds of those websites and their visitors to the attacks, Sekoia said. [3]
   • The hidden code first read each visitor's browser and device. [1] The attackers' server then sent the attack that matched. [1]
   • On iPhones running iOS 18.5, it was built to take WhatsApp and Telegram records, saved passwords, location and photos. [1]
   • On Android phones, it was built to take Chrome login data, browsing history and saved Wi-Fi passwords. [1]
   • The malware did not stay on the phone. [1] A restart stopped it until the phone visited an infected site again. [1]
   • One set of tools, DarkSword, was first documented in March and has been used by several hacking groups. [1]
   • Google researchers said customers of PARS Defense, a Turkish surveillance firm, have used DarkSword. [2]
   • Sekoia's analysis supports the idea that APT29 got these attacks from someone else. [3]
   • CISA, the US cyber-defence agency, added one of the flaws to its list of flaws under attack on 4 November. [5] Apple fixed that flaw in May 2024. [5]

   Who is involved

   • APT29 a hacking group widely linked to Russia's Foreign Intelligence Service; Sekoia ties these attacks to it
   • Sekoia a French security firm; it found the campaign and published the link to APT29
   • PARS Defense Turkish surveillance firm
   • Plerdy a Ukrainian tool that tracks website visits; its script was tampered with for a week
   • Apple and Google the makers of iPhone and Android software; both had already fixed every flaw used
   • CISA the US cyber-defence agency; it listed one of the flaws as under attack

   How it unfolded

   1. Mid-October Sekoia finds the campaign
   2. 4 Nov Sekoia publishes, and CISA lists the flaw [3][5]
   3. 27 Aug Plerdy's script is tampered with, for a week [3]
   4. 3 September the tampering ends
   5. May 2024 Apple fixes one of the flaws later used [5]
   6. 4 November CISA lists it as under attack
   7. Mid-Oct Sekoia finds the campaign [4]
   8. 18 Nov US federal agencies' deadline to patch [5]

   Why it matters

   Sekoia says several sites are likely to draw important visitors in the war between Russia and Ukraine. [4] Army, special forces and drone-equipment sites were among them. [4]

   A 2024 APT29 campaign on Mongolian government sites took cookies only for chosen websites. [3] This one took any cookie in the browser. [3]

   How it works

   Threat analysts call this a watering hole attack: the hackers poisoned websites and waited for visitors' phones to arrive. [1]

   The hidden code reads each phone's software version, so the server picks an attack that still works on it. [1]

   Seen it before

   A pickpocket waits at the busy station everyone passes through, rather than knocking on each door.

   The truth

   • The link to Russia rests on similar past attacks and overlapping online systems, Sekoia says. [1] It is a judgement of moderate to high confidence, not a proven fact. [3]
   • Who gains State hacking groups such as APT29 — Sekoia says packaged kits cut the cost and skill needed, so state hackers now reuse them. [4]
   • Nobody knows yet

     How many phones the Russian-linked attackers actually broke into.

     Sekoia did not say how many devices were compromised. [3]

   • Nobody knows yet

     Whether the attackers wanted particular people or anyone at all.

     Sekoia says it remains unclear whether they aimed at visitors to certain sites or took from as many old phones as they could. [4]

   Good news

   • Sekoia says a phone running up-to-date software is safe from these attacks. [3]

   What comes next

   US federal agencies have until 18 November to patch that flaw. [5]

1. 02

   French bank loses about $133 million to hackers

   What happened

   • Credit Mutuel Arkea, a big French banking group, confirmed on 4 November that hackers took money from it. [6] A crypto investigator puts the theft above $133 million. [6]
   • The bank would not confirm the figure, but said it could "validate the order of magnitude". [6]
   • The hackers went through the platform of a service provider, which the bank has not named. [6] That gave them direct access to the banking group. [6]
   • They then used the bank's own account to send money to other banks. [6] The bank says that money may then have been turned into crypto. [6]

   Who is involved

   That investigator, ZachXBT, works under a false name and first reported the theft on Monday. [6] The figure comes from tracing withdrawals from crypto exchange accounts. [6]

   • ZachXBT traces crypto, under a false name

   Why it matters

   The bank says customer accounts were not reached and customers' money is safe. [6] It says the loss is "absolutely not of a nature" to threaten its finances. [6]

   How it works

   Risk managers call this third-party risk: Credit Mutuel Arkea's attackers came through a supplier, not the bank's front door. [6]

   The supplier's platform could reach the bank, so taking over that platform carried the attackers inside, the bank says. [6]

   Seen it before

   Someone holding the office cleaner's keys walks straight past the front desk.

   The truth

   • Why it works on people Credit Mutuel Arkea says its systems spotted the attack very quickly, and that is reassuring to hear, but the bank gave no dates that would let anyone check it. [6]
   • ZachXBT says the attack ran over several days. [6] The bank says its systems spotted the activity very quickly. [6]
   • The bank also disputes French press reports that the way in was Fortuneo, its online bank. [6] It calls the attack sophisticated and not human error, and gives no further detail. [6]
   • Nobody knows yet

     How much Credit Mutuel Arkea really lost, and through which supplier.

     The bank says the final loss is not established, and it has not named the supplier. [6]

   Good news

   • Some of the converted money has been frozen, and the bank is working to recover the rest. [6]

1. 03

   Hackers set French blood stores to heat up

   What happened

   • Hackers changed settings on cold rooms at 3 sites of France's national blood service on 2 November. [8] The service, EFS, supplies blood to French hospitals. [8]
   • The rooms normally sit at minus 30C, and the setting reportedly went to 110C. [9] That does not mean the rooms or the blood reached that heat. [10]
   • The rooms hold frozen plasma for transfusions and blood samples. [9] EFS itself was not the direct target; its fridge supplier was. [8]
   • Shop fridges were also hit this week, in 8 European countries by one count and 10 by another. [12][9] France's health ministry says remote access to the blood fridges is off. [13]

   Who is involved

   Coolworld, a Dutch fridge rental firm, confirmed an incident with its remote monitoring software. [10] It says the flaw was in another supplier's part. [8]

   • Coolworld rents out cold rooms

   Why it matters

   Coop in Sweden has also cut its 7 hit shops' fridges off from the internet. [12]

   Ahold Delhaize, a Dutch supermarket group, expects October's attack to cut quarterly sales by up to 250 million euros. [13]

   How it works

   Security researchers call a decoy system a honeypot: 3 firms set up fake fridge software to watch these hackers. [11]

   Decoys run by Lupovis, a Scottish firm, saw probing 10 days before the second wave. [11] Only decoys placed in Europe were probed. [11]

   Seen it before

   A shop leaves a marked wallet on a table to learn who picks it up.

   Why the days an intruder stays decide how many people get hurt

   Incident responders call it dwell time: the longer an intruder stays unseen, the more people pass through what they can reach.

   The twist

   Two attacks can use the same way in. The one that stays longer reaches more people, because new customers keep arriving while it is there.

   The picture

   Plerdy tracker0.25 months

   NIKKE launcher1 month

   DreamHost2 months

   Amadeus37 months

   Step by step

   1. 1 An intruder gets in, and nothing breaks that anyone can see
   2. 2 Ordinary work carries on, so people keep logging in and booking
   3. 3 Each new person adds their details to what the intruder can read
   4. 4 The number of people hurt keeps growing until someone notices
   5. 5 Responders count back to the first day, and the total jumps

   The same force, elsewhere today

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

   • Amadeus break-in ran over 3 years

     On Japan Airlines' dates, new bookings kept entering the system from August 2023 to September 2026, so every passenger in those years joined what could be read.

   • DreamHost intruder stayed about 2 months

     Only people who logged in to the file manager during those months lost a password, so each week added more of them.

   • NIKKE launcher ran hidden code, a researcher says

     Every player who opened the launcher after 27 September ran the changed files, a researcher says, so each new launch could add a victim.

   • Hackers set French blood stores to heat up

     A break-in through a supplier was caught in one night, when a technician saw the temperature rise, and no blood was lost.

   Where you've seen this

   Food recalls

   a bad batch reaches more kitchens every day it stays on the shelf

   Disease outbreaks

   each day before the first case is spotted adds people to be traced

   A leaking water pipe

   the bill depends on how long the leak ran, not on how big the hole was

   The catch

   Dwell time is known only after the fact, and it counts from the first sign found, so the real first day may be earlier. A short stay can still do great harm: an investigator says Credit Mutuel Arkea's money went in days.

   And the whole of it

   If Japan Airlines' dates are right, a passenger who booked a flight in 2025 could not see that someone was already inside the firm their airline had chosen, and the airline could not see it either. Most of us sit in systems like this, and we hear about the break-in from a letter that counts backwards.

   The truth

   • What is really going on France's blood service was the opposite case, because a technician saw a fridge temperature rise overnight and no blood was lost. [10]
   • Russian hacktivist groups say they hacked Danfoss, a fridge-control maker. [11] Sekoia says it has no technical evidence that Russian-speaking groups were involved, and one group has made false claims before. [11]
   • Coolworld's statement, as reported, names neither the part nor the supplier it says was at fault. [8]
   • Nobody knows yet

     Who hacked the fridges at France's blood service and in Europe's shops.

     No formal blame has been made, and Sekoia's own report points to both Russian and Iranian actors. [11] The Register says it is unclear whether the blood stores were reached the same way as the shops. [9]

   Good news

   • France's blood service lost no blood or plasma after hackers set its cold rooms to heat up. An on-call technician saw the temperature rise overnight, and staff moved the stock.

1. 04

   Amadeus break-in ran over 3 years, Japan Airlines says

   What happened

   • At least 7 Asian airlines have now reported data thefts traced to Amadeus, a Spanish firm that runs airline bookings. [14][15]
   • 6 are Japan Airlines, Cathay Pacific, Korean Air, Asiana, Singapore Airlines and Malaysia Airlines. [14] Local media also reported All Nippon Airways. [15]
   • Japan Airlines counts 92,000 customers, Cathay Pacific 53,000 records and Singapore Airlines 46,000 records. [14]
   • Japan Airlines says Amadeus told it the access ran from 10 August 2023 to 23 September 2026. [15] Cathay Pacific gives those months for the bookings hit. [16]
   • Some records held passports, frequent flyer numbers and email addresses. [14] Amadeus says it saw no sign that payment data was reached. [14]

   How it unfolded

   Amadeus serves over 400 airlines and 210 airport operators. [15] It says it has fixed the flaw. [14]

   1. 10 August 2023 Access begins
   2. 23 September 2026 Access ends

   Why it matters

   Singapore Airlines will contact people with upcoming flights to change their exposed booking numbers. [14] It also warned customers to be wary of malicious emails. [14]

   How it works

   On Japan Airlines' dates, new bookings kept entering the system from August 2023 to September 2026, so every passenger in those years joined what could be read.

   The idea, in story 03: Why the days an intruder stays decide how many people get hurt →

   The truth

   • Amadeus told one airline the intruders used login details for a system account. [14] How they got those details has not been explained. [14]
   • Who pays Passengers on airlines using Amadeus — Singapore Airlines passengers with upcoming flights must change their exposed booking numbers, and it has told customers to be wary of malicious emails. [14]
   • Nobody knows yet

     How long the intruders were really inside Amadeus.

     Japan Airlines says access ran from August 2023 to September 2026, while Cathay Pacific gives those months for the bookings affected. [15][16]

   • Nobody knows yet

     How many people Amadeus's break-in reached in total.

     Amadeus declined to say how many victims there are, so the only counts come from each airline. [14]

2. 05

   White House internet addresses up for rent

   What happened

   • Anyone could rent internet addresses from the White House network for a few dollars, security firm Layer3 Intel found. [7]
   • Computers or phones of staff there seem to carry proxy software. [7] It lets strangers send their web traffic out through that device. [7]
   • Layer3 counted 236 exits through that network between 5 October and 3 November. [7]
   • The traffic used 6 addresses on 26 of 30 days. [7]
   • Layer3 found 28 separate proxy services could exit through this network. [7]

   Who is involved

   Synthient, another tracking firm, linked some to Windows 11 computers running Infatica, a proxy often bundled with free apps. [7]

   Infatica's founder is Vladimir Fomenko, a Russian. [7] His hosting firm, King Servers, held 6 of the 8 addresses Russia's spies used against US election databases in 2016. [7]

   • Synthient tracks proxy firms
   • Infatica proxy bundled with free apps
   • Vladimir Fomenko founder of Infatica

   Why it matters

   Spur, a tracking service, found such software on nearly 300 government networks and 318 utilities in January. [7]

   It also found it at 166 healthcare firms and 141 finance firms. [7]

   To a website, the traffic looks as if it comes from the White House network, not from the renter. [7]

   How it works

   Network security calls this a residential proxy: other people's traffic leaves through a real user's device and address. [7]

   App makers can embed proxy code and earn a small sum per install. [7] So a free app on a device brought to work can turn it into a rented exit. [7]

   Seen it before

   A stranger signs your name at the club door, and staff see only your name.

   The truth

   • Layer3 says it found this by chance, without aiming at the White House network. [7] The White House did not immediately reply to questions. [7]
   • Who gains Sellers of residential proxy access — Some resellers charge as little as 60 cents a gigabyte, and buyers can skip identity checks, Synthient's founder says. [7]
   • Nobody knows yet

     Whose devices at the White House were carrying proxy software, and whether anyone misused them.

     The tracking firms report the traffic and the software and name no person, and the White House had not replied. [7]

3. 06

   DreamHost intruder stayed about 2 months

   What happened

   • DreamHost, a US web host, says an intruder was inside its web file manager from August until 23 October. [17]
   • The intruder took upload logins and passwords from anyone who used that file manager in those months. [17]
   • Some webmail sessions may have been exposed too, DreamHost says. [17] TechCrunch reports hackers stole customer emails and credentials. [18]
   • A spokesperson says well under 10% of customers were hit. [18] DreamHost claims about 400,000 paying customers. [18]
   • DreamHost says it found no sign of access to control panel logins or card details. [17] It has closed the flaw and cancelled passwords and active sessions. [17]

   How it unfolded

   1. August Intruder gets in
   2. 23 October DreamHost finds it

   Why it matters

   DreamHost is telling those affected to check website files and email forwarding rules. [17]

   How it works

   Security engineers call this session hijacking: with a stolen DreamHost session, an intruder can act as that user. [18]

   A session token keeps someone logged in after the password check, so stealing it skips that check. [18]

   Seen it before

   A festival wristband gets you back in without showing your ticket again.

   Only people who logged in to the file manager during those months lost a password, so each week added more of them.

   The idea, in story 03: Why the days an intruder stays decide how many people get hurt →

   The truth

   • What is really going on Several of today's breaches were found long after they began: Japan Airlines says the access at Amadeus ran for just over 3 years, and DreamHost's intruder stayed about 2 months. [15][17]
   • DreamHost declined to say why it did not detect the attack until last week, citing the investigation. [18] It has not named the flaw. [18]
   • Who pays DreamHost's website owners — They are told to check their own website files and email forwarding rules for unauthorised changes. [17]
   • Nobody knows yet

     How many DreamHost customers lost their login details.

     DreamHost says only well under 10% of about 400,000, and left BleepingComputer's questions on the count unanswered. [17][18]

4. 07

   A fix that did not fix Zyxel switches

   What happened

   • Zyxel, a maker of network gear, has fixed a new flaw in its GS1900 switches, boxes that link computers. [19]
   • GreyNoise, a firm that watches attacks, counted 996 switches hacked across 48 countries from around 17 August. [19]
   • Then Binarly, a firm that studies code inside devices, found the hackers' tool in an exposed archive. [19] It still worked on all 10 firmware versions patched in June. [19]
   • Zyxel confirmed the new flaw on 11 October and released updates on 22 October. [19] It published its advisory on 3 November. [19]

   Who is involved

   The hackers took device settings and scrambled administrator passwords. [19] GreyNoise links them to a suspected Chinese-speaking group it calls Kapibala. [19]

   • Kapibala suspected Chinese-speaking group

   How it unfolded

   At first GreyNoise and CISA, the US cyber-defence agency, blamed an older flaw Zyxel had fixed in June. [19]

   1. June First fix
   2. 22 October Real fix
   3. 3 November Advisory

   Why it matters

   Switch owners who installed June's fix were still open to this tool until the October update. [19]

   How it works

   Software engineers call this an incomplete fix: June's Zyxel update added a check, and the attackers got past it. [19]

   The check trusted a counter that the attackers could change first, so the check passed. [19]

   Seen it before

   A door count looks fine if someone resets the clicker before the guard reads it.

   The truth

   • Zyxel classes this as a separate flaw. [19] It is unclear whether the tool was built to beat June's fix or used a related hole the fix missed. [19]
5. 08

   OpenAI's AI writes 948 Linux attacks in a test

   What happened

   • OpenAI says an upcoming AI model produced 948 working attacks on Linux in a 30-hour run in mid-October. [35]
   • Each lets someone already on a computer as an ordinary user take full control of it. [35]
   • The run looked at 2,308 possible flaws and reproduced 1,724 of them. [35]
   • It cost $1,213,430 in AI credits, or $1,279 per attack. [35]

   How it unfolded

   Human researchers review each attack and its fix before reporting it to the Linux maintainers, OpenAI says. [35] In June, its earlier model had found 24 such attacks. [35]

   Why it matters

   Google's threat team says flaws disclosed each month rose from 5,045 in January to 10,740 in August. [35]

   Britain's cyber agency says AI tools have almost certainly led to more unknown flaws being found. [35]

   How it works

   Security researchers call this exploit development: proving a flaw can really be used, as OpenAI's run did 948 times. [35]

   Each proof cost OpenAI $1,279 in AI credits. [35] VulnCheck, a firm that tracks flaws, says such tools already let novices find flaws with little skill. [35]

   Seen it before

   A metal detector finds coins for a treasure hunter and for a thief alike.

   The truth

   • These figures come from OpenAI's own report, and the model behind them is not yet released. [35]
6. 09

   Attack wipes servers at big Russian web host

   What happened

   • Beget, one of Russia's largest web hosts, says attackers erased data across its cloud services overnight on Sunday. [20]
   • They got into the systems used to manage its computers, then ran scripts built to erase data. [20]
   • All binary backups of its rented virtual servers were lost as well. [20] File-level backups survived and may recover some data. [20]
   • More than 1,000 servers must be checked by hand before they come back, Beget says. [20]

   Who is involved

   Beget ranks second among registrars of Russian .RU web addresses, with about 540,700, or 14.7%. [20]

   How it unfolded

   The European Union sanctioned Beget this year over services to Russian occupation authorities in Ukraine. [20]

   Why it matters

   Small and mid-sized firms that host online shops on Beget were mainly affected, the newspaper Kommersant reported. [20] Chestny Znak, Russia's product labelling system, was also disrupted. [20]

   How it works

   Disaster-recovery planners keep an offline backup, one copy no main system can touch; Beget lost all its binary backups. [20]

   Beget's file-level backups survived, and it says nearly all shared websites can likely be recovered. [20]

   Seen it before

   Keeping the only spare house key inside the house does not help when you are locked out.

   The truth

   • Beget found no sign that attackers copied customer data, but it tells customers to change passwords and keys anyway. [20]
   • Who gains Beget — It has told customers the attack may count as force majeure under Russian law, which could excuse it from some contract duties. [20]
   • Who pays Small Russian firms hosted on Beget — Their online shops depend on servers that must be checked one at a time, and some virtual server data may not return. [20]
   • Nobody knows yet

     Who wiped Beget's servers and how they got in.

     Beget has not said how the attackers entered, citing its investigation, and no group has claimed the attack. [20]

7. 10

   Fake nuclear emails from a hacked Ukrainian paper

   What happened

   • Hackers got into the website system and newsletter service of the Kyiv Independent, a Ukrainian English-language news outlet. [21]
   • They emailed false messages to subscribers in Russian and English. [21] One claimed President Zelensky had launched a nuclear warhead at Kyiv. [21]
   • Others made up an attack on Zelensky. [21] False posts also went on the website, and staff quickly removed them. [21]
   • Members' bank details were safe, it says, because outside services process payments. [21]

   Who is involved

   The outlet says it has nearly 3 million monthly readers and over 35,000 paying members. [21]

   How it unfolded

   It came days after hackers sent fake nuclear alerts through Ukraine's air raid app. [21] No link between the two is confirmed. [21]

   Why it matters

   The editor, Olga Rudenko, called it "an attempt to sow confusion and erode trust". [21]

   How it works

   Communication researchers call it source credibility: these hackers used a trusted outlet's newsletter to carry a false claim. [21]

   Subscribers had signed up to that newsletter, so the false email came from a name they already read. [21]

   Seen it before

   A prank note feels real when it arrives on the head teacher's own paper.

   The truth

   • The outlet blames suspected Russian actors because of what the messages said. [21] It gave no technical evidence for that. [21]
8. 11

   Flaws let one virtual computer take over a Xen host

   What happened

   • Xen is software that runs many virtual computers on one real machine. [34] Two new flaws let one of them read and write the others' memory. [34]
   • That includes dom0, the virtual computer that manages all the others. [34]
   • Every Xen version from 4.6 on is affected, the Xen Project said on 3 November. [34] There is no known workaround, only the patch. [34]
   • Qubes OS, a security-focused desktop system, says every supported setup is affected. [34] Debian still listed 3 of its releases as open on 4 November. [34]

   Who is involved

   Xen credits Anthropic, an AI company, with finding the flaws using its Claude model. [34] Ada Logics and CodeAnt AI also confirmed or reported them. [34]

   • Anthropic AI company credited with the find

   Why it matters

   XCP-ng, a platform built on Xen, says a single compromised guest can take over the whole host. [34]

   How it works

   Operating-systems courses call this isolation: each virtual computer may touch only its own memory, and Xen checks that. [34]

   To save time on a batch of copies, Xen reused checks it had already made. [34] Sometimes it did not check again when it should have. [34]

   Seen it before

   A guard checks the first badge in a group and waves the rest through.

   The truth

   • Xen 4.6 to 4.16 are affected, yet the advisory has no patch for them. [34]

   Good news

   • Xen's advisory mentions no attacks using these flaws, and patches came with it. [34]
9. 12

   Flaw in the tool that installs Linux software

   What happened

   • A flaw in APT, which installs software on Debian and Ubuntu Linux, lets an attacker swap a checked package. [33]
   • The attacker's code then runs with full control of the computer. [33]
   • The attacker must be able to change the download traffic, as a bad mirror or proxy could. [33]
   • It relies on APT retrying failed downloads, which it does by default from Debian 12 and Ubuntu 22.04. [33]
   • Debian, Ubuntu and APT's developers all published notices on 4 November. [33] Debian has released fixes, and Ubuntu is releasing its own. [33]

   How it unfolded

   1. 4 November Notices published
   2. Now Fixes going out

   Why it matters

   Ubuntu says systems built on Debian and Ubuntu are affected too, without naming them. [33] Older releases may also be open where a script has turned retries on. [33]

   How it works

   Security engineers call this a chain of trust: APT checks a package before installing, and everything rests on that check. [33]

   This flaw lets the attacker change a package after APT has checked it, which defeats the check. [33]

   Seen it before

   A teacher marks a test, and someone swaps the paper before it goes in the file.

   The truth

   • APT's developers call the flaw critical, while Ubuntu rates it high at 7.7 out of 10. [33]
   • The notices describe the attack over unencrypted web traffic and do not say whether encrypted sources are safe. [33]
10. 13

    Image upload flaw in the web toolkit Rails

    What happened

    • Ruby on Rails, a toolkit for building websites, has fixed a flaw rated 9.8 out of 10 for danger. [31]
    • A stranger who uploads a crafted image can write files on the server, with no login. [31]
    • Apps that keep uploads on their own disk can also have secret files read, including database passwords and outside logins. [32]
    • An app is hit only if 3 things hold. [31] It uses ImageMagick, an image tool, leaves its rules open, and takes strangers' uploads. [31]
    • Rails says the standard ImageMagick rules shipped by Debian and Ubuntu do not block this. [31]
    • Rails fixed it in new 8.0 and 8.1 versions. [31] Older release series are unsupported and will get no fix. [31]

    Why it matters

    Those secrets can open the other systems a site connects to, Dark Reading reports. [32]

    How it works

    Programmers call this unvalidated input: ImageMagick picks how to read a file from its contents, not its name. [32] Rails set no limit on which readers an upload could use. [31]

    So a file named like a picture can make the tool run a part that reads and writes other files. [32]

    Seen it before

    A parcel marked socks holds a note telling the post room to open the other parcels.

    The truth

    • Rails says one attack route was reported to it, but it does "not assume it is the only one". [31] Its advisory does not say whether anyone has used the flaw. [31]
11. 14

    One hacker, stolen app keys, 27 organisations

    What happened

    • A lone hacker used stolen app passwords to sign in to Microsoft cloud accounts at IT service firms. [23] Todyl, a security firm, reported it. [23]
    • The hacker tried 45 app logins at about 18 organisations and got into at least 33. [23]
    • That opened 27 organisations, including customers in energy, government and food. [23]
    • The hacker then downloaded 81 files and reached about 2,300 mailbox items. [23] Todyl says the attempts are continuing. [23]
    • About 95% of sign-ins came from scripts, with bursts of 24 sign-ins in 10 seconds. [23]

    How it unfolded

    Machine accounts can outnumber people by as many as 144 to 1, Entro Labs found by mid-2025. [24]

    Why it matters

    IT service firms run systems for many customers, so these stolen logins reached firms in other industries. [23]

    How it works

    Identity specialists call these non-human identities: accounts that software uses, not people. [24]

    Such accounts sign in with no person present. [23] Many have no owner, no second login check and no expiry date. [24]

    Seen it before

    A spare key under the mat never asks who is using it.

    The truth

    • The speed fits AI-assisted tools, but researchers have not tied the attacks to any one tool. [23]
12. 15

    Chinese keyboard app used to plant spy tools

    What happened

    • A China-aligned group hijacked Sogou Input Method, a popular Chinese-language keyboard app, to plant spy tools from 1 July. [22]
    • The app fetched some settings over an unencrypted connection. [22] The attackers redirected that request and sent back settings that installed the malware. [22]
    • Victims were in Hong Kong, Thailand, Japan, Vietnam, Sri Lanka and Oman. [22] They included a gambling firm, a Dubai internet cafe and a student in Denmark. [22]
    • Tencent, which owns Sogou, acknowledged the report on 21 September and fixed it in a version released in late October. [22]

    Who is involved

    ESET, a security firm, named the group SlippyNunchucks in research published on 4 November. [22] It says the group is tied to a state surveillance effort aligned with China. [22]

    • ESET security firm
    • SlippyNunchucks the group it named

    How it unfolded

    1. 21 September Tencent acknowledges it
    2. Late October New version

    Why it matters

    ESET says the number and spread of victims first made it suspect a software flaw. [22] They needed only to be running a keyboard app. [22]

    How it works

    Network engineers call this DNS hijacking: DNS turns a server's name into an address, here the attackers' address. [22]

    The request was not encrypted, so the app took a wrong server's answer as real. [22]

    Seen it before

    Someone swaps the sign at a road junction, and drivers follow it to the wrong town.

    The truth

    • ESET says the downloader part still does not check file types or limit where files are saved. [22]
    • Nobody knows yet

      Where the redirect that hijacked the Sogou keyboard app happened.

      ESET says it could be routers or an internet provider, and it does not know which. [22]

    Good news

    • Tencent moved the Sogou keyboard's traffic to encrypted connections in a version released in late October, after ESET reported the hijacking.
13. 16

    SonicWall will not fix its retired tool

    What happened

    • SonicWall, a maker of network security boxes, has warned of 2 flaws in GMS, its older tool for managing them. [36]
    • The first lets anyone with network access read internal files, with no login. [36]
    • The second lets someone with high-level access read or change its database. [36]
    • SonicWall says no fix will come, because GMS has reached end of life. [36] It urges customers to move to its newer tool, NSM. [36]

    Who is involved

    Researchers at Pentest-Tools found and reported the flaws. [36]

    Why it matters

    CISA, the US cyber-defence agency, has listed 19 SonicWall flaws as under attack in 4 years. [36] Ransomware gangs used 13 of them. [36]

    How it works

    Software makers call this end of life: after it, as SonicWall says of GMS, no more fixes come. [36]

    SonicWall will release no update, so these 2 flaws stay open on any box still running GMS. [36]

    Seen it before

    An old phone keeps working after its updates stop, so nothing looks wrong.

    The truth

    • Who pays Organisations still running SonicWall's GMS — They are told to move to the newer tool, or they keep 2 flaws that will not be fixed. [36]
14. 17

    IT worker admits passing US secrets to Ethiopia

    What happened

    • Abraham Teklu Lemma, 53, a US government IT worker, has pleaded guilty to sending classified files to Ethiopia. [28]
    • He agreed to a prison term of 120 to 262 months, the US Justice Department said on Tuesday. [28]
    • Court papers say he copied from at least 133 intelligence reports without permission to see them. [28]
    • Prosecutors say he burned files to discs, deleted their secret markings, and walked out with discs in his pocket. [28]
    • He also deposited more than $66,000 in banks during that time, prosecutors say. [28]

    Who is involved

    He worked for the US National Reconnaissance Office, State Department and Justice Department between 2019 and 2023. [28]

    • Abraham Teklu Lemma US government IT worker, 53
    • National Reconnaissance Office
    • State Department
    • Justice Department

    Why it matters

    Prosecutors say he sent the files to senior Ethiopian officials through an encrypted app. [28] An FBI official says they used them in military operations during a conflict. [28]

    How it works

    Counter-intelligence calls this the insider threat: Lemma's job let him move some files out of classified systems. [28]

    Clearance says a person may see secrets, and need-to-know says which ones. [28] Lemma had clearance, but no need-to-know for those 133 reports. [28]

    Seen it before

    A school caretaker has keys to every room, yet no right to read the exam papers.

    The truth

    • Court papers differ on when it began: the first complaint said July 2022, and prosecutors later said 2021. [28]
15. 18

    Canadian extradited over university supercomputers

    What happened

    • Ryan James Roach, 35, was extradited from Canada to the US on 28 October. [25] He appeared in a federal court in Rochester, New York, on Tuesday. [27]
    • Prosecutors say he and Mathiew James Stubbings, 41, hacked universities in 4 US states from 2016 to 2019. [25]
    • They allegedly stole logins and took over at least one supercomputer to mine cryptocurrency. [25] That left the owner with much less of its computing power. [25]
    • They allegedly hid behind rented servers in several countries, registered under false names. [26]

    How it unfolded

    1. 28 October Extradited
    2. Tuesday In court

    Why it matters

    A grand jury indicted both men in April 2022, over 2 years after the hacks are said to have ended. [25]

    How it works

    Security analysts call this cryptojacking: the alleged hackers ran mining software on computers the universities were paying for. [25]

    Mining turns computing power into coins, so the pair allegedly looked for universities that had supercomputers. [25]

    Seen it before

    Someone plugs a hidden charger into your socket, and you pay the bill.

    The truth

    • These are charges, not findings: neither man has been convicted. [25] Officials did not say how much crypto was mined. [25]

    What comes next

    Each faces up to 10 years in prison and a $250,000 fine. [26] The US is still seeking Stubbings's extradition from Canada. [27]

16. 19

    NIKKE launcher ran hidden code, a researcher says

    What happened

    • The makers of Goddess of Victory: NIKKE say its PC launcher files may have been tampered with. [29] The game has over 45 million downloads. [29]
    • A researcher called MotherSphere says 4 script files on the publisher's own delivery servers were changed on 27 September. [29]
    • The researcher says the scripts fetched a remote-access tool whenever a player opened the launcher. [29]
    • It could run commands, copy files, take screenshots, and steal game logins and browser data. [29]
    • Another part could reach crypto wallet recovery phrases, BleepingComputer found. [29] It is unknown whether attackers used it. [29]
    • The researcher says a tampered copy of 1 file was still being served after the update. [29]

    How it unfolded

    1. 27 September 4 files changed, the researcher says
    2. 4 November Client update goes out

    Why it matters

    The makers issued a client update on 4 November and have since posted steps to check a PC. [29] Those found infected are advised to reinstall Windows, BleepingComputer says. [29]

    How it works

    Software security calls this a supply-chain attack: the NIKKE launcher ran whatever its publisher's delivery servers handed it. [29]

    Changing the files at that one source reached every player who opened the launcher afterwards. [29]

    Seen it before

    Tampering at the bottling plant reaches every shop, without anyone visiting a shop.

    Every player who opened the launcher after 27 September ran the changed files, a researcher says, so each new launch could add a victim.

    The idea, in story 03: Why the days an intruder stays decide how many people get hurt →

    The truth

    • Level Infinite, the publisher, had not answered BleepingComputer about the researcher's findings at publication. [29]
    • Who pays NIKKE players whose PCs were infected — BleepingComputer's advice is to reinstall Windows and change every password from a clean device. [29]
17. 20

    Apple will check public iCloud links

    What happened

    • Apple says it will check items shared through public iCloud links for "harmful content", from version 27.3 of its systems. [30]
    • This covers the "Anyone with the Link" option, not items shared directly with named people. [30]
    • Apple says it may remove such content and block accounts. [30] It also reserves the right to involve the police. [30]
    • Version 27.3 reached developers as a test on Tuesday, and a general release is expected in December. [30]
    • Public links already drop end-to-end encryption, even with Apple's stricter data setting on. [30] Apple says it can access those items. [30]

    How it unfolded

    1. Tuesday Test version
    2. December General release expected

    Why it matters

    The change reaches everyone who shares a file by public link, on iPhones, iPads and Macs. [30]

    How it works

    Online platforms call this content moderation: Apple will now judge what people publish through public links. [30]

    A public link removes end-to-end encryption, the lock only sender and receiver can open, so Apple can check the file. [30]

    Seen it before

    A sealed letter stays private, but a poster on the school wall can be read by the caretaker.

    The truth

    • Apple's help page does not define harmful content. [30] It also does not say whether people or software will decide. [30]
18. 21

    Calendar invites used to steal Google logins

    What happened

    • A phishing campaign stole live Google logins from company staff from mid-August, Material Security reports. [38] Phishing means tricking someone into handing over a login. [38]
    • The messages arrive as Google Calendar invites, often from an organisation the recipient already knows. [38] They pose as voicemails, court summonses or missed payments. [38]
    • Most links point first to Google's own address, so a link checker sees nothing wrong. [38]
    • A hidden server then sits in front of the real Google sign-in page. [38] Codes and phone prompts pass straight through to Google. [38]
    • Material says 897 web addresses were taken down in a month. [38] It says the campaign now looks finished, but rebuilding it would be cheap. [38]

    Why it matters

    A live session reaches a person's mail and Drive files, and can grant other apps access in their name. [38]

    How it works

    Security engineers call this a reverse proxy: the attacker's server passes everything between the victim and the real Google. [38]

    The victim approves a real prompt and the operator keeps the logged-in session, so a second check does not help. [38]

    Seen it before

    A fake cloakroom attendant hands your ticket to the real desk, and keeps your coat.

    The truth

    • This comes from a sponsored post written by Material Security, which sells a product against this. [38] It is the only account we could read. [38]
    • Who gains Material Security — Its report on the Calendar campaign also says its own product caught over 99% of the messages. [38]
19. 22

    US cyber-sharing law runs out on 11 December

    What happened

    • Tuesday's vote gave both houses of the US Congress to Democrats. [40] Those houses hold the committees that oversee CISA, the US cyber-defence agency. [40]

    Who is involved

    Maggie Hassan is in line to lead the US Senate Homeland Security committee. [40] Bennie Thompson is poised to return as House Homeland Security chair. [40]

    • Maggie Hassan Senate Homeland Security
    • Bennie Thompson House Homeland Security

    How it unfolded

    CISA has cut staff by nearly one-third and is looking to hire again. [40] Thompson says his first aim is to speed up that re-hiring. [40]

    1. 2015 Law passes
    2. 11 December Law expires

    Why it matters

    The US Senate's defence bill has a very broad permission for offensive hacking run by contractors, a policy expert says. [40]

    How it works

    Lawmakers call this a sunset clause: the 2015 law ends on 11 December unless US lawmakers renew it. [40]

    Rand Paul, the current US Senate committee chair, has been the biggest obstacle to renewal, a former official says. [40] Two experts say what he does next could decide cyber bills. [40]

    Seen it before

    A library card stops working on a set date unless someone renews it.

    The truth

    • Hassan says the Trump administration has been "gutting CISA". [40] That is a rival party's charge, in a statement to CyberScoop. [40]

    What comes next

    A 2015 US law on sharing cyber-threat information is set to expire on 11 December. [40]

    Funding for CISA's parent department runs out the same day. [40] Lawmakers could still extend both before the new members arrive. [40]

20. 23

    Chrome 157 fixes 213 flaws

    What happened

    • Google has released Chrome 157 with fixes for 213 flaws in the browser. [37] 7 of them are rated critical and 77 high. [37]
    • The most common kind, at 51, is a missing or wrong permission check. [37]
    • Google has paid $31,000 in rewards so far. [37] It has not yet set the rewards for 39 of the flaws. [37]
    • Google mentions no attacks using any of these flaws. [37] The update is rolling out for Windows, macOS and Linux. [37]

    Who is involved

    Google, Anthropic's Xinyang Ge and the researcher Wenbin Liu reported the 5 critical flaws with no listed reward. [37]

    • Google
    • Anthropic
    • Wenbin Liu

    Why it matters

    Until a browser updates, those 213 flaws stay open on it, so SecurityWeek urges a quick update. [37]

    How it works

    Programmers call 45 of Chrome 157's flaws use-after-free: code that keeps using memory after giving it back. [37]

    Seen it before

    You keep using a locker after handing back its key, and someone else's things are now inside.

21. 24

    Meta pays $1.2 million for Muse AI bugs

    What happened

    • Meta has paid about $1.2 million in bug rewards for Muse, its personal AI agent. [41] Muse acts for a user across their email, calendar and messages. [41]
    • Of more than 630 reports since April, 64 were valid. [41]
    • The top reward, $300,000, went for a flaw found 4 months before launch. [41] It exposed conversations and tokens on machines Meta's staff used for testing. [41]
    • Meta took that service offline, fixed it and stopped treating the local machine as trusted. [41]
    • Another $90,000 went for a web page that tricked Muse when asked to summarise it. [41] Meta says its detection now catches that attack and 100 variants. [41]

    Why it matters

    Muse can act in a person's email and accounts, so a tricked agent acts in their name. [41]

    How it works

    AI security calls that prompt injection: words on a web page got treated as instructions by Muse. [41]

    The page copied status messages from Muse's own browsing tool, so Muse took them as real and acted. [41]

    Seen it before

    A note in the teacher's handwriting tells the class to go home, and the class goes.

    The truth

    • Meta says its top reward is several times what other AI developers advertise. [41] That comparison, and the 100 variants, are Meta's own claims. [41]

    Good news

    • A researcher found a flaw in Meta's Muse AI agent 4 months before it launched. Meta paid $300,000 and fixed it before the public could use the agent.
22. 25

    Vercel ties logins to the laptop

    What happened

    • Vercel, a cloud firm that hosts websites and AI agents, is rolling out new protection for logins to its platform. [39]
    • It ties a login session to one device, using a private key stored in the computer's security chip. [39]
    • So a stolen session cookie cannot be reused on another device. [39] A cookie here is the file that keeps someone logged in. [39]
    • "The private key can never be retrieved ever again," said Vercel's chief technology officer, Malte Ubl. [39]
    • It works in Chrome on macOS and Windows for now. [39] Vercel is rolling it out slowly to catch problems with some connected apps. [39]

    Why it matters

    Information-stealing malware copies cookies from browsers, then attackers log in as the victim. [39] Developers who manage their projects on Vercel are the people this protects. [39]

    How it works

    Security engineers call this device-bound credentials: a login that works only with a key the device keeps. [39]

    Vercel says malware can copy a cookie but cannot pull the key from the chip, so the copy fails elsewhere. [39]

    Seen it before

    A bus pass with your photo is no use to the person who steals it.

    The truth

    • This is Vercel's own description, and the rollout covers logins to Vercel's site, not its customers' sites. [39]

    Good news

    • Vercel has begun tying logins to a security chip in the user's computer. A stolen login cookie then fails on any other machine.
23. 26

    Survey: staff connect AI apps without asking

    What happened

    • DNSFilter, a security seller, surveyed 400 IT and security workers about AI at their companies. [43]
    • Three-quarters said their budgets had grown over the past 12 months. [43]
    • 46% were confident that unapproved AI apps were connected to company systems. [43]
    • One-fifth said an unapproved AI link had already led to a security incident. [43]
    • 40% said their next dollar would go on AI security, ahead of cloud security at 25%. [43]

    Why it matters

    DNSFilter says connecting an AI tool to company data is often decided by someone who does not own the risk. [43]

    How it works

    Security teams call this shadow AI: AI tools staff connect without approval, as in this survey. [43]

    37% said they learn of a connected AI app only after the fact, so no check runs before it joins. [43]

    Seen it before

    A pupil brings a friend into a locked club without telling the teacher who holds the list.

    The truth

    • These figures come from a firm that sells protection against this problem. [43]
24. 27

    Investors give AI data-security firm $150 million

    What happened

    • Sentra, a security firm in New York and Tel Aviv, has raised $150 million, led by Capital One Ventures. [42]
    • Sentra's software finds sensitive company data across cloud and office systems. [42] It maps what people, AI agents and service accounts can reach. [42]
    • It says it will spend on its AI agent tools, plus sales, marketing and new regions. [42]

    How it unfolded

    That takes its total raised to over $260 million. [42] It was founded in 2021. [42]

    Why it matters

    Its chief executive says old network defences were not built for AI agents moving at machine speed. [42]

    How it works

    Security engineers call this least privilege: Sentra's product gives each user or AI agent only the access its job needs. [42]

    Seen it before

    A babysitter gets the house key, not the code to the safe.

    The truth

    • What the software does is the firm's own account, from its funding announcement. [42]