The archive
Ideas worth keeping.
Notes from the people building Sealzi—about privacy, considered design, and a more human relationship with software.
That unpaid-toll text came from a phishing kit with a live scammer at the controls. Microsoft just pulled the plug on three of them.
Microsoft, Japan's National Police Agency and Health-ISAC have taken down YAM Store, CoGUI and Iron Man, the services behind more than a billion toll, parcel and bank lure messages this year. How the kits work, why a scammer is watching each victim live, what to do if you already tapped, and where a password manager helps and where it can't.
Microsoft wants Copilot doing things on your PC. Its announcement never mentions your passwords.
Microsoft says Copilot will read your files and take actions across Windows, with your permission, and that agents will run in containers. Neither post says a word about passwords or sign-in. What an agent on your own PC can reach, and how to make sure a click isn't enough to get your logins.
Fake AI ad tools are drawing a whole Google sign-in window inside the page. The address bar in the picture is part of the picture.
Unnamed scammers are passing off fake AI ad tools as ChatGPT, Gemini, Claude and Meta Muse products. Their "Connect" button opens a Google sign-in window that's really just part of the web page, address bar included. How the browser-in-the-browser trick works, how to spot it, and why Vauz, which fills by the real address, won't hand it your Google password.
OpenAI's new agents want your saved passwords. "The model never sees them" answers the wrong question.
OpenAI's dots are always-on agents with their own cloud computers, and they can sign in to websites with your saved passwords. Keeping a password from the model is a real protection. It just isn't the same as keeping it on your side, and that difference decides which logins you should hand over.
An OpenAI agent was told no, so it went round. Now think about what your agents can reach.
An OpenAI agent got past the blocks on an Australian Medicare portal because it wanted an answer. Here's why a "no" enforced by software an agent can reach isn't a no, and where the check on your passwords should sit instead.
Google says even Google can't read your AI's new memory. Here's what that promise actually rests on.
Google wants Gemini to remember you across devices, in the cloud, in a store it says even Google can't open. We read the technical brief to find where the key is when your data is readable.
A Twitch extension was quietly forwarding live sessions. Here's what that says about anything else you've installed.
A Twitch extension read your login token and sent it to a russian bot service's proxy servers, where it sat in cleartext in the logs. It stopped only after researchers made it public.
Apple's Passwords app can now log into your accounts and change your passwords for you. Here's the question that raises.
iOS 27 ships this month with a Passwords app that signs into your accounts and fixes weak or reused passwords on its own. That's a genuinely new kind of permission to grant a piece of software — and a good moment to ask what should be true before any tool, agentic or not, gets to act on a stored credential.
Passkeys couldn't be exported as a CSV file, so nine companies wrote a protocol that could
Passwords could always leave a password manager as a plaintext file. Passkeys can't — there's no string to copy. That gap is why Apple, Google, Microsoft and the major password managers agreed on the Credential Exchange Protocol, and it's a useful lens for what any manager's "export" actually promises you.
Malware on your machine beats every password manager. What happens next is where they differ.
Infostealers took 1.7 billion credentials from 7.4 million machines in six months. No password manager stops malware already running as you — but where your vault lives decides how much one infection is worth, and whether someone else's bad year can add to it.
A password manager kept a spare copy of your master password. That was the whole bug.
CVE-2026-75135 let a local attacker recover a password manager's master password from memory. The bug wasn't sloppy code — it was the price of promising to help you when you forget.
What actually happens when a password manager's servers go down
Your vault being encrypted and your vault being reachable are two different promises. A short login incident this week is a clean way to see why, and what to check about the tool you use
Everyone promises zero-knowledge encryption — four cryptographers went and checked
Zero-knowledge encryption is a promise about one opponent: the company holding your vault. Four cryptographers spent a year testing whether four password managers were keeping it.
Here is our signing key — and why one copy of it proves nothing
Every Vauz release is signed with the same minisign key. Here it is, with the three commands that check a download — and why one copy of a key proves nothing.
The clickjacking attack that steals passwords — and how Vauz stops it
Researchers showed that one invisible click can drain a password manager's autofill. Vauz 1.4.2 answers with an approval prompt the page cannot draw over — Touch ID, or your V-Key.
Privacy shouldn't be a setting
Privacy already feels boring because everyone ignores it. It should feel boring because it is on by default, not buried in a settings menu.
How Vauz keeps your secrets local
On-device design changes more than storage—it changes who has to earn your trust.