Every security check passed last week. That was the problem.
Last week the security checks all did their jobs. A backup app compared its update against the checksum it was given, and it matched. A video app asked its update server for news, and the server answered. Certificates came with perfectly valid padlocks. Websites hashed passwords exactly the way their security pages promise. Capcom looked for leaked customer passwords and couldn’t confirm a single one. Every box got ticked, and every one of those stories still went badly, because the trouble was always somewhere the check wasn’t looking
So this Monday, instead of one story, here’s the week, sorted by which check passed and why it didn’t help
AOMEI BackupperUpdate check
Popcorn TimeUpdater
eMudhraCertificates
php2shellPassword storage
Capcom IDBreach notice
php2shell: patched on Wednesday, working by Saturday
On Wednesday PHP shipped an emergency release for two bugs, nicknamed php2shell, that let a complete stranger run code on a web server without signing in to anything. We wrote it up the same day, because of where it lands: at the one moment your password is plain text, when it has just arrived at the server and the site hasn’t hashed it yet. The hashing still happens, exactly as promised. It just happens one step too late, after the attacker’s code has already run
At the time, a few security vendors said they were seeing attacks and several said they weren’t. By Saturday afternoon that argument was over. Honeypots run by Adel Karimi, who leads detection and response at Perplexity, recorded 35 confirmed break-ins between 15:35 and 18:57 UTC on October 24, each attack request followed by a shell starting up inside the web server. The decoys ran PHP 8.4.26, one release short of the fix. Karimi is careful to say he built the favourable setup on purpose, and that it proves nothing about how reliable the attack is anywhere else. Expel’s Marcus Hutchins called it the best confirmation of real-world exploitation he’d seen so far. No real victim has been named yet, and CISA’s list of exploited bugs still doesn’t include either one
Then on Friday PHP published a note about a third bug , CVE-2026-115560, with no fix yet. PHP’s own description is that the setup it needs is “unlikely in practice”, and the workaround (switching off session upload progress) is one that the php2shell advisory already recommended. Three bugs in one week. PHP is having the kind of October usually reserved for a horror franchise 🎃
What to do: if you run a PHP site, update to 8.5.12, 8.4.27, 8.3.36 or 8.2.35 today. If you only log in to PHP sites, which is everyone: when a site you use says it was hit, assume the password you typed there is exposed, because an attacker running code on the server sees it as it arrives, before any hashing. Don’t change it on that site until the site says it’s fixed, or your new password walks out the same door. Do change it today anywhere else you used it 🚪
eMudhra: a padlock signed with the number 4
On Saturday the 17th, a researcher told Mozilla that eight intermediate certificates from eMudhra, an Indian certificate authority, had secret keys that were “integers less than 256” (bug 2083449 ). Not 256 bits. Under 256. Somebody published one of them, and it was 4 . (No, not 4 secret keys, the secret key was just number “4”. Fun, yeah? 🙂) Those intermediates could vouch for any website at all, and browsers and operating systems had trusted them for about eighteen months. Your bike lock has more combinations
Browsers mostly had it covered. Mozilla blocked the certificates in under three hours, without anyone needing to restart Firefox, and Chrome blocked them the same day. According to the incident tracker (open it, it runs checks for you) run by Mozilla’s Dennis Jackson, Microsoft followed on the 20th and Apple on the 22nd. Browsers that insist on Certificate Transparency would have refused the unlogged certificates anyway. The rest of your computer is less tidy
This morning, on one of our Macs, Apple’s own networking, which Safari and most Mac apps go through, refused all four (4 again xD) of the tracker’s test sites. The Mac’s built-in curl and Python connected to every one of them, on Friday and again this morning, including the two sites a browser would have refused for never being logged. Those tools skip Apple’s checks and read a certificate file that ships with macOS, and that file still trusts the roots that vouch for the broken certificates. The tracker says Apple plans to update the file in a future macOS release. Until then, the built-in command-line tools on a Mac will accept a padlock signed with the number 4 🙆♂️
Browsers and systems, as reported
- Oct 17Firefox and ChromeBlocked the day of the report
- Oct 20WindowsBlocked
- Oct 22AppleBlocked
Our Mac, tested
- blocked
- blocked
- blocked
- blocked
- 200 OK
- 200 OK
- 200 OK
- 200 OK
That doesn’t mean anyone attacked anyone. eMudhra says it has seen no sign the keys were abused, which is a promise from the company whose keys were a single digit, so we’ll wait for the final report
What to do: update your browser and your operating system. If you use developer tools, update those too. The tracker lists fixed versions of curl’s certificate bundle, Python’s certifi, Node.js, Go and Electron, and marks Homebrew as patched. The built-in tools on a Mac have to wait for Apple. The tracker does offer a manual workaround that edits the system certificate file, so read it before you run it, and leave work machines to whoever manages them
AOMEI Backupper: the checksum came from the attacker
Google’s threat intelligence group and Mandiant published this one on the 15th, and ManageEngine, whose software caught it first, followed the next day . Then on Monday AOMEI shipped its answer, so it counts as last week’s. An attacker Google tracks as UNC7017 edited the files on AOMEI’s own update servers between April and July, left each edit up for five to ten minutes, then put the real files back. Confirmed windows: May 28, July 16 and July 20, and Google thinks there were probably more. A copy of Backupper that checked in during a window was sent a backdoor. One that checked a minute later got the real update, and so did every security scanner that came to look. A smash-and-grab with a five-minute parking meter
The update check itself worked fine. Backupper reads a small file from AOMEI’s server that says where to download the package and which MD5 checksum it should match, then compares the two. The attacker edited that same file, so they changed both, and Backupper checked the attacker’s package against the attacker’s number and found a perfect match. Homework marked by the student who copied it 🙄 A second edited file, the one that drives Backupper’s promotional pop-ups, then launched what had been unpacked. The downloader was picky about its victims, too: it quit on machines in China and on machines running certain security tools
July 20 came with a bonus: an invisible one-pixel window titled “Flash Sale” that makes Windows try to sign in to the attacker’s server, handing over a scrambled proof of your Windows password that can be cracked offline. The flash sale was your password
Here’s the part that ages badly. ManageEngine told AOMEI on June 7 . AOMEI’s first theory was that the files had been changed in transit, and two of the three confirmed windows came after that warning. On Monday, October 19, AOMEI released Backupper 8.5.1 , and its statement says it has switched off the in-app pop-up and automatic update channels, dropped the download domain named in the reports, and added “asymmetric encryption protection” to the configuration files. That might mean a signature and might mean something else, and AOMEI doesn’t say whether the download itself is now checked against one. On Tuesday it told The Record it still hadn’t identified or notified the customers who were infected. Still working out who got the flash sale 🤷♂️
What to do: if you’ve had Backupper installed and updating at any point between April and July, update to 8.5.1, then check for leftovers. Open Task Scheduler and look for tasks named RadminComponentRuntime or NitroRuntime, and paste %LOCALAPPDATA% into File Explorer’s address bar to look for a Radmin or Nitro Pro folder you didn’t create. AOMEI’s AnyViewer uses the same update system, according to ManageEngine, so check that machine too. If anything turns up, treat the PC as somebody else’s. A backdoor with a remote shell can do anything you can do on that machine, and no password manager stops malware already running as you, ours included. Back up your documents, reinstall Windows, and change your passwords from a different, clean device, starting with your email, because email resets everything else
Popcorn Time: an updater that would run anything for anyone
On Thursday, Eye Security’s Niels Teusink told the story of a security firm quietly fixing millions of strangers’ computers, which is not a sentence you get to write often. The torrent-streaming app Popcorn Time, in its old Time4Popcorn version, has an updater that asks a list of web addresses whether there’s a new version, then runs whatever program the answer names, with full administrator rights, over plain HTTP and with no code signing. Its developers seem to have lost control of those addresses long ago. Eye spotted the updater on a client’s laptop calling one that nobody owned any more, registered it, and started getting about 50 requests a second
Eye’s best estimate is 2.5 to 3 million Windows machines and 250,000 to 350,000 Macs. Anyone who registered those addresses could have run anything they liked on all of them. Eye used them for exactly one thing: between November 2025 and March 2026 it served an update that does nothing except switch the updater off, then handed the 15 addresses it owned to the Shadowserver Foundation to be locked away. Some of the others still belong to people Eye doesn’t know. Its reward was eleven antivirus vendors flagging its update as malicious within a week. No good deed goes unflagged 🍿
The check here wasn’t wrong, it was missing. The updater checked that somebody answered, and never asked who
What to do: uninstall it. Eye’s update only disabled the updater, and the rest of the app still loads code from abandoned addresses, which is how Eye made a calculator pop up on its own test machine. Your films will survive
Capcom: no passwords leaked, 196 changed anyway
Capcom says someone got into CAPCOM ID, the account you need to play some of its games online. It spotted the intrusion on October 19, and the haul was 11,774,247 records: email addresses, plus when each account was created, when it last signed in, how many times, and from which IP address. Forty-three of those records also had a birthday, gender, country and nickname. Credit card details weren’t included
On passwords, Capcom chose its words carefully. It “has not confirmed any compromise” of passwords that customers set themselves. It has confirmed that 196 accounts had their passwords changed without permission, and on Friday it added that for 19 of those, the intruder went on to open the customer’s page at e-Capcom, its Japanese online store, which holds things like name, address, phone number, date of birth and purchase history. No leak confirmed, and also changed by a stranger. Schrödinger’s password 🐈
Capcom hasn’t said how those 196 passwords got changed, and neither will we. What it does warn about is phishing: nearly 12 million email records tied to a games account is exactly the list someone needs for convincing fake Capcom emails, texts and calls
What to do: treat any message about your CAPCOM ID as suspicious, and get to Capcom’s site by typing the address rather than clicking a link. Turn on two-factor authentication, which CAPCOM ID offers with an authenticator app from your account page. And if the password you used there also works somewhere else, change it there first
The quick ones
Chrome 156 arrived on Tuesday with 353 security fixes, six of them rated Critical. Ten were in the parts of Chrome that handle saved passwords, autofill and sign-in credentials, one rated High. Google hasn’t said any were used in attacks. The fixes only take effect once Chrome restarts, so when it offers to relaunch, let it. It’s been asking
Hijacked Tor exits went after SSH passwords. Between October 1 and 5, more than 300 Tor exit relays sent SSH traffic through interception servers, showing users a changed server key. Anyone who clicked past the warning handed over their server password, and the attacker logged in to install a backdoor. The Tor Project threw them off the network . The write-up landed on Tuesday, and its advice is the old advice: log in with keys, not passwords, and never click through a changed-key warning
Spyware that reads your Telegram password before Telegram can hash it. A researcher took apart uCreator , a rent-a-spyware platform for Android that ships as repackaged copies of real apps like Telegram, WeChat and GCash, installed from outside the app store. Its Telegram module is built to grab the two-step verification password in plain text before the app hashes it. Same lesson as php2shell, in your pocket: hashing protects a password after the moment someone else’s code could have read it. Install messaging apps from the official store, and nothing else
Where Vauz fits, and where it doesn’t
Two of last week’s lessons are ones we build around
A signature, not just a checksum. A checksum tells you what a file should be. A signature tells you who said so, and only the holder of a secret key can say it. Every Vauz release is signed with minisign, and the secret key isn’t on our web server or in any repository. So someone who took over our download page could swap the installer and the checksum list, AOMEI-style, but couldn’t make a signature that matches. We ran the check again this morning: the real file passes, and the same file with one character changed fails
Checked withOur public key. Save a copy and check every download against itRWTcejyNskGe1Xc8GvEbKjUlI2wPh8XLVcs8Xcd3EraJvmSX7LuX8QQY
The catch is where your copy of the key comes from, which is why our signing-key post asks you to save it once and check every release against your own copy, not against whatever a page shows that day. It’s three commands, and no certificate authority is involved, not even the ones that can count past 4
A separate login per site. php2shell and Capcom are both stories where the website is the one in trouble and you’re just a customer, and nothing on your computer stops a break-in somewhere else. What Vauz changes is what that break-in is worth. It keeps each site’s login separately, in a vault on your own computer, and hands one over only to the exact site it was saved for, after you’ve approved that site yourself in a prompt from the Vauz app. A site that gets popped learns the password you used there, and that password opens nothing else
And where it doesn’t fit: if a backdoor like AOMEI’s is already running on your PC, no password manager helps, ours included. Clean the machine first
What to watch this week
Whether a real php2shell victim gets named, and whether CISA finally lists it. When Apple updates the certificate file behind the Mac’s built-in tools, and whether eMudhra’s final report explains how a “test” produced keys that small. Whether AOMEI says how the attacker got into its update servers, and whether “asymmetric encryption protection” turns out to mean a real signature on the download. Whether Capcom explains the 196. And the fix for PHP’s third bug, which PHP says will come with its next release
It’s Monday. Go and install the update that’s been sitting in your corner. We’re doing ours the moment this goes up 🗓️
Signed, and separate
Vauz signs every release and keeps a separate login per site.
Every Vauz release is signed with a key that isn't on our download server, so a swapped installer fails the check. Your logins stay in a vault on your own computer, each site's separately, so a breached website only ever learns the password you used there. The free plan stays completely free for life, with Plus and Premium available when you need more!
Use Vauz completely free — for, like, ever