Ok, I think I am getting close. So while SE keys are not actually stored inside the Secure Enclave keys generated by the Security framework are (or can be, depending on whether kSecAttrIsPermanentis set or not) stored in the keychain. Keys generated by Apple CryptoKit are not stored at all and there is no way to convert between the two kinds of keys. Correct?

Thanks Quinn! You know, there are days where I feel I would not get anything done without you helping me out... Does that mean the dataRepresentation of a SecureEnclave.P256.Signing.PrivateKey is already stored in the keychain by the system, but is inaccessible to client code so the only way for it to get a reference to such a key is to store the dataRepresentation itself and recreate the key from it? And if the key had instead been created by not using CryptoKit, but the Security framework (passing .privateKeyUsage in the access control list) it could later be retrieved by SecItemCopyMatching, correct?

Why do you need this in nscurl specifically? Mostly just to understand how specific status codes are handled on iOS. To elaborate: I have been hitting the same issue described here and here - the URL Loading System raises an error if it receives a response with the status code 403 and a client certificate was involved in the handshake (regardless of whether the certificate was accepted or not). I (re-)raised this as FB10026351. In my case all requests to the server require client authentication and the server may return 403 for unrelated reasons (e.g. for domain errors). The server team agreed to use a different status code for those cases but wanted me to make sure they don't pick one where a similar situation could arise. In my lab appointment the network engineer pointed out that there are in fact status codes other than 403 that receive special treatment and suggested using nscurl to quickly check how specific status codes are handled, but apparently the behaviour changes if a client certificate is involved. Here is what I found: Without client authentication: If no client certificate is involved and the server responds with 401 the client gets an error (NSURLErrorCannotConnectToHost) If no client certificate is involved and the server responds with 403 the client gets the response With client authentication: If a (valid) client certificate is involved and the server responds with 401 the client gets the response If a (valid) client certificate is involved and the server responds with 403 the client gets an error (NSURLErrorClientCertificateRequired) However, the scenarios involving a client certificate cannot be tested with nscurl, hence my question.

Thanks Quinn! The Random Unification - https://github.com/apple/swift-evolution/blob/master/proposals/0202-random-unification.md Swift evolution proposal makes the same assumption. Without a definitive source one could also assume that arc4random always utilizes RC4 and conclude that it (and thus the Swift SystemRandomNumberGenerator - https://developer.apple.com/documentation/swift/systemrandomnumbergenerator) is to be avoided, though.

For the record: It seems that the problem was caused by xcodebuild not waiting for the pre-action script to complete before proceeding with the build. In that case the entitlements file was not present at the location pointed to by CODE_SIGN_ENTITLEMENTS (it is in fact not under version control there, it is instead copied there by the pre-action script and being gitignored afterwards). A reboot of the Jenkins node was enough to remedy the problem - at least for the time being. The CSSM errors mentioned above seem to be unrelated after all.

I have been watching the logs closely during on of the failing builds and I see codesign report a "UNIX error exception: 17" (EEXIST maybe?) followed by several CSSM errors (e.g. CSSMERR_DL_DATASTORE_DOESNOT_EXIST) before xcodebuild reports a "UNIX error exception: 22" (EINVAL?). Does that ring any bells?