Two flaws this week came from defaults trusting input that nobody had checked
Two security write-ups published on 5 November describe different systems and the same pattern: a component trusts a value because of where it sits, not because anyone checked it.
In the first, Lauri Koskela showed that Rails, Jetty and Spring, in common configurations, give the Forwarded header priority over the X-Forwarded headers a reverse proxy writes. Proxies usually rewrite the X-Forwarded family and leave Forwarded alone, so a client can put a false IP address or host in it and the application reads it as if the proxy had vouched for it. The trusted-proxy list that guards against this checks where the request came from, and the request really did come from the proxy.
In the second, Alpine Linux found that sshd, built without PAM, compares the output of crypt() with the stored password field. musl's crypt() returns an asterisk on a key longer than 256 bytes, and an asterisk is what the field holds for every account without a password. A long enough password therefore matches any key-only account, as long as one other account on the machine has a password hash.
Neither project had an obvious bug in isolation. The proxy forwards what it does not understand, the framework follows a published standard, the C library returns its documented error value and sshd compares strings. The failures happened at the joins, where one side assumed the other had already done the checking.
The fixes follow the same line. Spring Boot 4.2 now reads only X-Forwarded headers by default and Rails 8.1.4.1 extended its spoofing check; Alpine shipped patched packages for every branch from 3.16, and the immediate mitigation is to turn password logins off.

What it means
Both cases argue for deciding explicitly which input a component accepts, rather than inheriting whatever a neighbour lets through. A value is trustworthy only if something along the path actually set or checked it.