It was found that the krb5_read_message() function of MIT Kerberos did not correctly sanitize input, and could create invalid krb5_data objects. A remote, unauthenticated attacker could use this flaw to crash a Kerberos child process via a specially crafted request.