What stops a copy of 1Password working?

This is the one category where a subtle mistake is not a bug but a breach, and you will not find out from a stack trace. Key derivation parameters, secure random generation, constant-time comparison, memory zeroing, and clipboard hygiene are each individually easy to get almost right. Sharing a secret between two people without either the server or the transport ever seeing plaintext requires real cryptographic design, not a library call. Recovery is the genuinely hard problem: users forget the passphrase, and any mechanism you build to help them is also the mechanism an attacker will use. Autofill must resist phishing by matching origin correctly, which is a surprisingly deep rabbit hole involving public suffix lists.

Can AI copy

1Password

Q: Can AI copy 1Password?

Nice try. The moat was never the code. Realistically you won't.

A vault for the secrets that unlock everything else you own.

Nice try

Verdict: Nice try
Time to copy: you won't
Category: ⚿ Security & identity
Source: Proprietary

The prompt

Written to be pasted straight into an agent with no editing. It describes the whole job, including the parts that will go wrong.

Do not roll your own crypto, and be careful what you build here at all. If you want a personal vault, use an audited open-source manager and stop. If you are learning the domain: derive a key from a passphrase with a modern memory-hard KDF at recommended parameters, encrypt each item individually with authenticated encryption, and ensure the server only ever stores ciphertext it cannot decrypt. Generate secrets with the platform CSPRNG only. Match autofill against the registrable domain using the public suffix list so a lookalike host cannot claim a credential. Have someone competent review it before it holds anything real.

Open inClaude ↗Terminal agents take it from the clipboard.

What copies cleanly

The interface. A searchable list of items, a generator, a detail view, an autofill prompt.
Encrypting a blob with a key derived from a passphrase, which is a handful of lines against a good library.
Browser extension autofill and a sync-a-file-somewhere model.

Where it stops

This is the one category where a subtle mistake is not a bug but a breach, and you will not find out from a stack trace.
Key derivation parameters, secure random generation, constant-time comparison, memory zeroing, and clipboard hygiene are each individually easy to get almost right.
Sharing a secret between two people without either the server or the transport ever seeing plaintext requires real cryptographic design, not a library call.
Recovery is the genuinely hard problem: users forget the passphrase, and any mechanism you build to help them is also the mechanism an attacker will use.
Autofill must resist phishing by matching origin correctly, which is a surprisingly deep rabbit hole involving public suffix lists.

What actually protects them

Audited cryptography and a published security model, which is trust you accumulate rather than write.
Nobody sensible adopts a password manager from an unknown author, and they are right not to.
Being the thing enterprises already approved.

Editor's note

The only entry on the index where the honest advice is to not do it. Everywhere else a mediocre copy is merely disappointing; here it is a liability you have handed to someone who trusted you.

Somebody already did it

Go and look at the original

We would rather you formed your own opinion than took ours. They built the thing; we only measured it.

1password.com ↗