TWIL October 23rd 2026

There are a wholeeee lot of edge cases to consider with OIDC SSO, for example:

  1. Attacker creates an account at an IdP with an (unverified) email they don't own but know the victim uses
  2. Victim tries to create account, realises it exists and resets password/logs in with some kind of 2FA
  3. Attacker can still log in via OIDC because they control the account at the IdP

The fun thing is this also works without OIDC if you allow users to create accounts without verifying their email and offer alternative login methods.

The spec says validating the ID token from an OIDC flow is optional if you're getting it straight from the IdP. This saves wayyy more complexity and effort than you might expect.

You still need a way to track users agreeing to your ToS in an OIDC flow, having the links on the IdP's consent screen is probably not sufficient.

Sadly Josh seems to have lost the motivation to write with AI just stealing all his work, but he's briefly back to write about some CSS features I might use for work in 5 years or so.