It is easy to read a specification and conclude the problem is solved. It is harder to sit with the twenty implementations that each read the same paragraph differently and still call themselves compliant.
This piece walks through how we think about it at Paycux, what we have changed our minds about, and where the sharp edges are.
The setup: How JWT verification is supposed to work
The setup: How JWT verification is supposed to work is where this gets concrete. Normalization is the whole job. Two providers can be perfectly compliant and still disagree about what a group membership is, whether a logout is meaningful, and which attribute carries the email address.
Treat anything the other side sends as untrusted input until it has been through validation you wrote. Signature checks, issuer checks, audience checks and expiry checks are four separate decisions, and skipping any one of them is a real vulnerability rather than a theoretical one.
The classic attack: RS256 to HS256
That brings us to the classic attack: rs256 to hs256. Treat anything the other side sends as untrusted input until it has been through validation you wrote. Signature checks, issuer checks, audience checks and expiry checks are four separate decisions, and skipping any one of them is a real vulnerability rather than a theoretical one.
Treat anything the other side sends as untrusted input until it has been through validation you wrote. Signature checks, issuer checks, audience checks and expiry checks are four separate decisions, and skipping any one of them is a real vulnerability rather than a theoretical one.
- Validate signature, issuer, audience and expiry as four separate checks
- Normalize provider attributes into one internal shape
- Fail loudly when a peer stops meeting a documented assumption
- Keep a fixture per provider so regressions surface in CI
Why the math works out
Why the math works out deserves its own treatment. Normalization is the whole job. Two providers can be perfectly compliant and still disagree about what a group membership is, whether a logout is meaningful, and which attribute carries the email address.
Version the assumption, not just the code. Write down what you expect from the peer, and make the failure loud when it stops being true, because the alternative is a connection that half works for six weeks.
Beyond the classic: related confusion vectors
Consider beyond the classic: related confusion vectors. Treat anything the other side sends as untrusted input until it has been through validation you wrote. Signature checks, issuer checks, audience checks and expiry checks are four separate decisions, and skipping any one of them is a real vulnerability rather than a theoretical one.
Normalization is the whole job. Two providers can be perfectly compliant and still disagree about what a group membership is, whether a logout is meaningful, and which attribute carries the email address.
Normalization is the whole job.
The alg: none attack
Consider the alg: none attack. Version the assumption, not just the code. Write down what you expect from the peer, and make the failure loud when it stops being true, because the alternative is a connection that half works for six weeks.
Normalization is the whole job. Two providers can be perfectly compliant and still disagree about what a group membership is, whether a logout is meaningful, and which attribute carries the email address.
JWKS injection via jku and x5u
That brings us to jwks injection via jku and x5u. Version the assumption, not just the code. Write down what you expect from the peer, and make the failure loud when it stops being true, because the alternative is a connection that half works for six weeks.
Version the assumption, not just the code. Write down what you expect from the peer, and make the failure loud when it stops being true, because the alternative is a connection that half works for six weeks.
Cross-algorithm confusion in asymmetric families
Cross-algorithm confusion in asymmetric families is where this gets concrete. Normalization is the whole job. Two providers can be perfectly compliant and still disagree about what a group membership is, whether a logout is meaningful, and which attribute carries the email address.
Normalization is the whole job. Two providers can be perfectly compliant and still disagree about what a group membership is, whether a logout is meaningful, and which attribute carries the email address.
Where this leaves us
The pattern repeats across every system we have looked at: the hard part is not the mechanism, it is keeping the mechanism honest as the surrounding assumptions change.
If you are working through the same problem and want to compare notes, the docs cover the mechanics and the console shows the behaviour on your own data.
Everything here, already built
Sign-in, enterprise SSO, directory provisioning, roles and an audit trail behind one API. Start with the quickstart and have a working sign-in this afternoon.