Day Lila
Day Lila

Tuesday, 22 September 2026

22 Sept · Cyber · In brief

Cyber · Tuesday, 22 September 2026

Cyber, in brief

23 stories · 27 sources

Swipe up for the day in brief ↑Scroll down for the day in brief ↓

What is moving · 1 of 4

AI doing the breaking in

Building ↑

Spain's privacy regulator logged its first breach report blaming an AI agent, which found a way in and then changed personal records. [13] Researchers showed that an ordinary browser add-on can take over the AI helper built into five popular browsers. [7] A security firm counted five ways an AI helper can be pushed into spending with no limit. [26]

What is moving · 2 of 4

Old kit still carrying the load

High →

A US watchdog said the two systems carrying text messages between controllers and airliners were built before encryption was normal. [3] Ransomware gangs joined the attacks on a VMware flaw patched back in July. [12] Three flaws in the Linux kernel, the core of most servers, were added to the US must-patch list. [20]

What is moving · 3 of 4

Gangs turning on each other

Building ↑

ShinyHunters took over the leak site of Clop, a rival extortion gang, and demanded money from it. [4] It threatened to publish which companies paid Clop and how much. [5] RansomHouse publicly accused Namibia's defence ministry of hiding a break-in. [9]

What is moving · 4 of 4

Days of work lost to attacks

High ↑

Ransomware at manufacturers is 40% up on the same period last year, by one security firm's count. [24] An insurer's survey put the average downtime after an incident at 32.8 hours. [25] Munich's biggest university took servers offline and briefly stopped enrolment. [8]

Part of the day's signal

One pressure seen across the day runs through Cyber too.

Google is fined 403 million euros over how it handled people's location data

Swipe up for the story ↑Scroll down for the story ↓

Ireland's Data Protection Commission fined Google 403 million euros, or $463 million, over the way it handled people's location data. [1][2]

it examined Google's conduct from 2018, when the law took effect, to February 2020 [1]

it said so on the day it announced the fine [1]

Who is involved

Who is involved

Why it matters

Watch the three other Google investigations the same regulator says are open, because whether any of them reaches a fine is the test of how fast one office can work for 27 countries.

What's really going on

A regulator that can take money took 403 million euros off Google, for location settings the company stopped defending in 2019.

Why it works on people

A number this large reads like the end of a story. The conduct it punishes ran from 2018 to February 2020, and Google says it changed how location data works from 2019.

Who gains

The case covers only the period up to February 2020, so the six years of location handling since it changed the settings are not judged here. [1]

Who pays

The regulator found the Location Accuracy setting inside Android was not handled lawfully, fairly or openly. [2]

Nobody knows yet

How many people the Irish fine actually covers.

The regulator published no count of affected users, and Google did not give one either. [1][2]

Part of a signal

Fines for what companies did years ago

Also seen in Tech.

The whole signal →

The short version

That's the story in short.

The whole story has 10 more parts.

or swipe up for the next storyor scroll on for the next story ↓

The whole story, in 10 parts ↓

Part 1 of 10 · Who else is involved

Part 2 of 10 · How it unfolded

Part 3 of 10

The regulator found Google did not handle that data lawfully or fairly in Web and App Activity, the setting that records what a person searches and visits. [1]

Part 4 of 10

It made the same finding about Location History, the service that maps the places a phone has been. [1]

Part 5 of 10

It also found Google was not lawful, fair or open about Location Accuracy, a setting inside the Android phone software. [2]

Part 6 of 10

Google said the case is about old policies and that it has changed how location data works since 2019. [1]

Part 7 of 10

Ireland decides these cases for all 27 European Union countries, because Google's European headquarters is in Dublin. [1]

Part 8 of 10

The investigation opened six years ago and examined the period from 2018, when the law took effect, to February 2020. [1]

Part 9 of 10

It is the fourth largest privacy fine this regulator has issued; a 1.2 billion euro fine on Meta, which owns Facebook and Instagram, is larger. [1]

Part 10 of 10

The regulator says three more privacy investigations into Google are still open. [1]

Watchdog says messages to airliners can be faked

Swipe up for the story ↑Scroll down for the story ↓

The US Government Accountability Office, a watchdog that audits federal agencies, reported on Monday that the two systems carrying text messages between air traffic controllers and airliners were built before encryption was normal and still have none. [3]

It said the US aviation regulator has not finished its risk assessments and cannot spot all of these threats as they happen. [3]

Both message systems were built before encryption was normal, so neither one seals what it sends.

Someone able to transmit on those channels could send a plane a fake cancellation of its clearance. [3] The regulator agreed with all nine recommendations. [3]

Why it matters

Senator Ron Wyden, who asked for the report, called the technology incredibly insecure. [3]

Why it matters

Jamming and faking of satellite navigation is already a problem over northern Europe, where Estonia and Finland blame Russia and Russia denies it. [3]

What's really going on

A watchdog that can only write recommendations told the US aviation regulator that messages sent to airliners can be faked, and the reply was that it agreed with all nine.

Nobody knows yet

Whether anyone has ever faked a message to an airliner.

The watchdog said the US aviation regulator cannot detect all of these threats as they happen, so no count exists to publish. [3]

One extortion gang now bills another

Swipe up for the story ↑Scroll down for the story ↓

ShinyHunters, a data-theft crew, took over the dark web leak site of Clop, a ransomware gang, and has spent the week demanding money from it. [4]

It first asked for an eight-figure sum, then for everything Clop earned from its attacks on Oracle's business software, with interest. [4]

On Monday it added a public apology to the list and said the price rises every 24 hours. [4][5] It also named three people it says run Clop. [5]

It claims to hold private keys and server data used to run Clop's operation. [6]

Why it matters

ShinyHunters has threatened to publish which companies paid Clop, how much, and the Bitcoin addresses used. [5] That threat lands on the victims, not on the gang.

Who gains

Holding Clop's leak site lets it demand a share of the Oracle campaign money, and neither crew can take the other to any court. [4][5]

Who pays

ShinyHunters says it will publish which firms paid, how much, and the Bitcoin addresses used. [5]

Nobody knows yet

Whether ShinyHunters really holds Clop's record of who paid.

The claim comes from the crew that defaced Clop's site, and Clop has said nothing in public; two researchers told Reuters the fight looks genuine. [4][5]

A browser add-on can seize the browser's AI helper

Swipe up for the story ↑Scroll down for the story ↓

Researchers at Forever Security found a way for an ordinary browser extension to cross into the AI assistant built into the browser, an attack they call BragJack. [7]

It worked on five of the most used AI browsers, from Google, Microsoft, Opera, Anthropic and Perplexity. [7]

In Chrome they could take screenshots, read files on the machine and switch on the camera and microphone with no click from the user. [7] All five companies have fixed it. [7]

Why it matters

Anyone running one of those browsers with a single extension installed was exposed, which the researchers put at hundreds of millions of people. [7]

Why it matters

The bounties paid ran from $600 to $7,000. [7]

Good news

Five browser makers - Google, Microsoft, Opera, Anthropic and Perplexity - each closed the hole that let an ordinary add-on take over the AI helper built into their browsers. Google and Microsoft gave the flaw a numbered record, and the researchers who found it were paid.

Munich's biggest university pulls servers offline

Swipe up for the story ↑Scroll down for the story ↓

Ludwig Maximilian University in Munich, which has more than 52,000 students, detected an intruder on its network on Wednesday and disconnected the affected server. [8]

The records that may have been taken include names, dates of birth, contact details, bank account information, health insurance numbers and the reasons some students asked for leave. [8]

The university has not said how many people are affected or when the access began. [8] Enrolment was briefly suspended. [8]

Why it matters

A university holds bank details and health identifiers for tens of thousands of people who never chose its security.

Why it matters

Some students say they still cannot register for courses or see their grades. [8]

Who pays

Names, dates of birth, bank account details and health insurance numbers may have been taken, and some students still cannot register for courses. [8]

Nobody knows yet

What was taken from Munich's biggest university.

The university has not said how many people are affected, when the access began, or how long it lasted. [8]

Good news

Munich's biggest university extended the enrolment deadlines it missed while its systems were offline, so students who could not register on time are not held back a term.

Ransom gang names Namibia's defence ministry

Swipe up for the story ↑Scroll down for the story ↓

RansomHouse, an extortion group active since 2021, listed Namibia's Ministry of Defence and Veterans Affairs on its leak site and accused that ministry's IT department of hiding the break-in. [9]

Namibia's national cyber incident team then confirmed unauthorised activity on the defence ministry's network. [9]

It has not said what was taken, whether files were encrypted, or whether a ransom was demanded. [9] Its head urged other organisations in the country to report attacks promptly. [9]

Why it matters

The country's own responders learned the shape of the incident from the criminals' website. They have no reach at all over the group itself.

How it works

Where a company keeps its office decides who can fine it

How it works

Google's European headquarters is in Dublin, so one Irish regulator decides what happens to the location data of people in all 27 European Union countries.

How it works · step 1 of 5

A company serves many countries from one European office

How it works · step 2 of 5

The privacy law sends every case to the country that office sits in

How it works · step 3 of 5

So one regulator decides for all of them

How it works · step 4 of 5

That regulator's speed becomes everyone's speed

How it works · step 5 of 5

Six years passed before this fine arrived

The twist

Choosing where to put a building is not usually thought of as a legal move.

The twist

For a company serving 27 countries from one office, it decides which country's regulator will ever get to judge it.

The picture

1 of 27

European Union countries, and the one whose regulator leads Google's privacy cases

Google's European headquarters is in Dublin, so Ireland's regulator handles the case for every country on this grid.

Seen the same day

The Canadian province suing OpenAI

British Columbia filed its case in California, because that is where OpenAI decided not to tell police about the account, not where the shooting happened

Seen the same day

Four countries acting on North Korean IT workers

Argentina and Pakistan opened cases against a money mover and a document forger living there, because no court can reach the workers' own government

Seen the same day

Namibia's defence ministry

Namibia's national cyber team can help that ministry recover and has no reach over RansomHouse, which operates from somewhere else entirely

The catch

One regulator is also the reason this fine is 403 million euros. Twenty-seven separate cases would each have been smaller, and most of them would still be running.

The whole

Every app on a phone answers to a regulator in one country, and the company picked that country when it opened an office there.

The whole

Nobody who uses the app was in that room, and neither were the 27 governments now waiting on one office in Dublin.

Who pays

A gang published a note accusing its IT department of hiding the break-in, and the national cyber team then confirmed unauthorised activity. [9]

Nobody knows yet

Whether Namibia's defence ministry lost any files.

The national cyber team has not said which systems were hit, whether anything was encrypted, or whether a ransom was demanded. [9]

New malware hidden in film torrents

Swipe up for the story ↑Scroll down for the story ↓

Kaspersky's research team said on Monday that a public torrent archive was compromised and used to hand out a previously unknown malware disguised as popular films, including The Odyssey. [10]

The first stage checks whether it is being examined by security software before it does anything. [10] It then takes administrator rights without the usual Windows warning and opens remote access. [10]

It finds its control server through the Solana blockchain, which makes the network harder to shut down. [10]

Why it matters

Several hundred victims have been found in Russia, Turkey, Japan, Kenya, Uganda, Colombia and several European countries. [10] They include companies, government bodies, retailers and transport firms. [10]

Who pays

Several hundred victims have been identified in Russia, Turkey, Japan, Kenya, Uganda, Colombia and several European countries. [10]

Check Point fixes the box that runs its firewalls

Swipe up for the story ↑Scroll down for the story ↓

Check Point, an Israeli security company, published a fix on 16 September for a flaw in its Security Management Server, the machine that holds firewall rules and administrator accounts. [11]

The flaw is rated 9.8 out of 10 and can be triggered before anyone logs in, by sending a login request with a very long username. [11]

A successful attack runs the attacker's own code with full control of the server. [11] Customers on automatic updates were already protected. [11]

Why it matters

The US cyber-defence agency recorded exploitation as none, and the scanning firm Censys found no public attack code. [11] This is the rarer case where the fix arrives before the attacks.

Good news

Check Point published a fix for a flaw rated 9.8 out of 10 in the server that holds its firewall rules. The US cyber-defence agency recorded exploitation as none, and customers on automatic updates were already protected.

Ransomware gangs join attacks on VMware

Swipe up for the story ↑Scroll down for the story ↓

The US cyber-defence agency said ransomware gangs have joined the attacks on a flaw in VMware vCenter, the software companies use to run their virtual servers. [12]

Broadcom, which owns VMware, patched it on 29 July and told customers to treat it as an emergency. [12]

Two weeks later an incident response firm found more than 361 addresses across 47 countries compromised through it. [12] About 450 vCenter servers are still reachable from the internet. [12]

Counts from two different groups. An incident response firm found the compromised addresses, in 47 countries; a monitoring group counts the servers still reachable online.

Why it matters

The fix has existed for nearly two months. What changed this week is who is using the hole.

Who gains

A VMware flaw patched on 29 July is still reachable on about 450 servers, and the US agency now lists it as used in ransomware attacks. [12]

Nobody knows yet

How many of the exposed VMware servers are still unpatched.

A monitoring group counts about 450 reachable from the internet, and cannot see which of them took the July fix. [12]

Spain logs its first breach blamed on an AI agent

Swipe up for the story ↑Scroll down for the story ↓

Spain's data protection regulator said it has received the first breach report in which an AI agent is blamed for carrying out the attack. [13]

The agency said the attacker used a well known large language model to find a way in, then changed personal records and opened invoices. [13]

It did not name the model, the organisation or the attacker. [13] It stressed that neither the model nor its provider was itself compromised. [13]

Why it matters

The agency's point was that one attacker ran several stages of a break-in with very little human work. [13][14]

Why it matters

This is a first report, which is not the same as a first event.

Nobody knows yet

Which language model was used in Spain's first AI-agent breach.

The Spanish regulator would not name the model, the organisation or the attacker, and says the report is still under review. [13][14]

A Canadian province sues OpenAI for a shooter's chats

Swipe up for the story ↑Scroll down for the story ↓

British Columbia, a Canadian province, sued OpenAI on Monday. [15]

The case is over the company's failure to tell police about a user's ChatGPT activity before the Tumbler Ridge school shooting in February, in which eight people died, six of them children. [15]

Reports say OpenAI's safety team flagged the account for references to gun violence months earlier. [15] The company's chief executive apologised to the families in April. [15]

The province filed in California, because that is where the decision not to report was made. [15]

Why it matters

British Columbia has asked for the chats and OpenAI has refused. [15]

Why it matters

Families of the victims have filed their own lawsuits, and any damages the province wins would go towards rebuilding the school. [15]

How it works

The Canadian province suing OpenAI

British Columbia filed its case in California, because that is where OpenAI decided not to tell police about the account, not where the shooting happened

The idea, in story 6: Where a company keeps its office decides who can fine it →

Nobody knows yet

What OpenAI's records of the Tumbler Ridge shooter say.

British Columbia's attorney general says she has not read them, and that OpenAI refused to hand them over. [15]

Apple Maps blurs thousands of sites in China

Swipe up for the story ↑Scroll down for the story ↓

Researchers outside China noticed in September that satellite pictures of military and government sites across the country had been blurred on Apple Maps. [16]

The BBC's verification team went through the platform and identified thousands of obscured locations. [16] Several of them are detention facilities linked to alleged human rights abuses. [16]

Why it matters

Satellite pictures are what outside researchers have used to document those sites. Blurring them on one widely used map takes away a record that anybody could look at.

Who gains

Apple Maps now blurs thousands of military and government sites in China, including detention centres researchers had documented from those pictures. [16]

Four countries act on North Korean IT workers

Swipe up for the story ↑Scroll down for the story ↓

A United Nations panel reported on Wednesday that Vietnam, Laos, Pakistan and Argentina have taken real steps against schemes that get North Koreans hired into well paid technology jobs using stolen or bought identities. [17]

A 140-page United Nations report last October described the scheme and said the workers live illegally in China and about 40 other countries. [17]

Argentina opened an investigation into a woman accused of laundering their earnings and froze some assets. [17] Pakistan opened a case against an alleged forger of identity papers. [17]

Why it matters

None of these cases touches North Korea itself. Each country is prosecuting the local person who made the scheme work where they live.

How it works

Four countries acting on North Korean IT workers

Argentina and Pakistan opened cases against a money mover and a document forger living there, because no court can reach the workers' own government

The idea, in story 6: Where a company keeps its office decides who can fine it →

Good news

Argentina opened an investigation and froze assets over money earned by North Koreans hired into IT jobs under stolen identities. Pakistan opened a legal case against an alleged forger who supplied them with identity documents.

Belgian sports bodies lose member records

Swipe up for the story ↑Scroll down for the story ↓

Belgium's French-speaking table tennis federation and its gymnastics federation were both hit by attacks disclosed this week. [18]

What the attacker claims to hold. The federation says its own checks have not yet confirmed any of it.

A person using the handle Venus1337 claimed responsibility and posted samples said to include names, membership numbers and records of sporting sanctions. [18] The claim covers about 1.69 gigabytes. [18]

The table tennis federation says its checks so far show no evidence its own systems were compromised, and it is looking at the supplier that runs its website. [18]

Why it matters

A small membership body holds years of personal records and buys its website from an outside firm. The break-in can happen at a company the members have never heard of.

Who pays

An attacker posted samples said to cover more than 66,800 members and 17,400 users, including records of sporting sanctions. [18]

200,000 WordPress sites run a flawed calendar

Swipe up for the story ↑Scroll down for the story ↓

How many websites use the add-on, and how many are still on a version with both flaws.

Two critical flaws in The Events Calendar, a WordPress add-on installed on more than 600,000 websites, can let a stranger run code and take over the site. [19]

Both are rated 9.8 out of 10 and neither needs a login; one of them works through the comment box on an event page. [19]

StellarWP, which makes the add-on, fixed one on 25 August and the other on 10 September. [19] WordPress figures show about 240,000 sites still run versions old enough to carry both. [19]

Why it matters

A fix that exists is not a fix that is installed. Every one of those sites has to update itself, and most are run by somebody with another job.

Good news

StellarWP published fixes for both critical flaws in The Events Calendar, a WordPress add-on used by more than 600,000 websites. The first went out on 25 August and the second on 10 September.

Three Linux flaws join the must-patch list

Swipe up for the story ↑Scroll down for the story ↓

The US cyber-defence agency added three flaws in the Linux kernel, the core of the software running most servers, to its catalogue of bugs known to be under attack. [20]

The most serious is rated 9.8 out of 10 and sits in the code handling encrypted network traffic, where it can crash a machine or leak memory. [20]

The other two can corrupt memory or produce wrong results from encryption work. [20] Federal agencies were given three days to patch all three. [20]

Why it matters

That agency can order US federal bodies and can only advise everyone else. The same kernel runs most of the machines behind the internet.

Fake password-manager installers kill 145 security tools

Swipe up for the story ↑Scroll down for the story ↓

LastPass, which makes a password manager, found a fake version of its authenticator app handed out through GitHub pages tuned to appear high in search results. [21]

The download installs a kernel driver, attested by Microsoft, built to shut down 145 antivirus and security products, and then runs a data stealer called Rapuncel. [21]

The download poses as a trusted brand, and what it installs switches off the tools that would have caught it.

LastPass says none of its own systems were touched, and that the campaign impersonates at least 40 companies. [21] It has been running for several months. [21]

Why it matters

The lure is a security product, which is exactly what a careful person goes looking for. Searching for the official app is how victims arrive.

Who gains

They impersonate at least 40 companies and ship a Microsoft-attested driver built to shut down 145 security products. [21]

Malware installs its own browser add-on

Swipe up for the story ↑Scroll down for the story ↓

Researchers at Elastic described malware they call KREMLIN, which waits for Chrome or Edge to close, copies an extension straight into the browser's profile folder and registers it as approved. [22]

To do that it takes the browser's own encryption keys and rebuilds the checks the browser uses to spot tampering. [22] The extension then poses as a media tool. [22]

It steals cookies and session data and logs what is typed into forms, including passwords. [22]

Why it matters

The owner never approves the add-on, and the browser treats it as though they had. [22] The extensions list is the one place it shows.

A backdoor disguised as four Windows tasks

Swipe up for the story ↑Scroll down for the story ↓

Securonix researchers described a campaign they call TASK#STOMP, which drops a script with a random file name on the victim's desktop and uses it to install a PowerShell backdoor. [23]

The backdoor collects business documents, watches for new files as they are created, takes screenshots, and steals Wi-Fi passwords and whatever is on the clipboard. [23]

It stays alive through four scheduled tasks named to look like Windows services, such as Network Audio Service. [23] It also alters file timestamps. [23]

Why it matters

How the first script reaches the desktop is not known, and the researchers think an emailed lure is likely. [23]

Why it matters

The altered timestamps are there to mislead whoever examines the machine afterwards.

Factory ransomware up 40% on last year

Swipe up for the story ↑Scroll down for the story ↓

A report from the security firm Black Kite says ransomware incidents at manufacturers this year are 40% above the same period last year. [24]

It argues that mid-sized manufacturers absorb most of the attacks, because they are the suppliers the larger firms build their products from. [24]

Its example is the attack that shut Jaguar Land Rover's UK plants in September 2025, which affected more than 5,000 other companies. [24]

The UK's Cyber Monitoring Centre put that one incident at 1.9 billion pounds. [24]

Why it matters

This is a security vendor's count of its own data, not an official figure. The direction is worth more than the exact number.

The average attack costs $52,000 and 33 hours

Swipe up for the story ↑Scroll down for the story ↓

The Hiscox Cyber Readiness Report 2026, published on 15 September, puts the average cost of a cyber incident over the past year at about $52,000, with average downtime of 32.8 hours. [25]

The highest average cost was in Italy, at $134,138. [25]

Firms based in the UK were the most likely to report a successful attack, at 38%, and firms in the US the least likely, at 20%. [25]

The share of firms in each country reporting a successful attack in the past year, from a survey of 6,800 security decision-makers.

Among victims, 69% reported staff burnout, high stress or a worse workplace afterwards. [25]

Why it matters

The numbers come from a survey of 6,800 people responsible for security at firms in the UK, Europe and the US, so read them as that company's own count. [25]

Why it matters

The 33 hours is the part an ordinary customer feels.

An AI helper can be made to spend without stopping

Swipe up for the story ↑Scroll down for the story ↓

A report from the security firm Forcepoint sets out five ways an AI system can be pushed into using far more computing than anyone budgeted for, a pattern it calls denial of wallet. [26]

The service stays up while the bill climbs, so nothing looks broken. [26] No single request has to look suspicious or break a usage limit. [26]

The industry body OWASP now ranks this sixth in its 2026 list of the main risks in AI applications. [26]

Why it matters

A misconfigured automation or a long session can do it with nobody attacking at all. [26] The damage arrives as an invoice rather than as an outage.

CISA says its own tool programme is too slow

Swipe up for the story ↑Scroll down for the story ↓

Richard Grabowski, who runs service delivery for the US cyber-defence agency's Continuous Diagnostics and Mitigation programme, told a conference on 15 September that it has to get faster. [27]

The programme supplies security tools and capabilities to other US federal agencies. [27] He said the way agencies worked together was not fast enough for yesterday's threats, let alone tomorrow's. [27]

His answer is automation at scale, so specialists spend their time on new threats rather than routine alerts. [27]

Why it matters

This is the same agency that hands federal bodies three-day deadlines to patch. It is saying its own side of the work moves too slowly.

The truth, across the day

Swipe up for the truth ↑Scroll down for the truth ↓

Who gains

The 40% rise at manufacturers is Black Kite's own figure [24] and the $52,000 average is Hiscox's own survey [25], and both are now the numbers being quoted.

That's the whole Cyber day.

23 stories, 27 sources.