SSH 1 through 3, and possibly other versions, allows local users to bypass restricted shells such as rbash or rksh by uploading a script to a world-writeable directory, then executing that script to gain normal shell access.

The SSH protocols 1 and 2 (aka SSH-2) as implemented in OpenSSH and other packages have various weaknesses which can allow a remote attacker to obtain the following information via sniffing: (1) password lengths or ranges of lengths, which simplifies brute force password guessing, (2) whether RSA or DSA authentication is being used, (3) the number of authorized_keys in RSA authentication, or (4) the lengths of shell commands.

CORE SDI SSH1 CRC-32 compensation attack detector allows remote attackers to execute arbitrary commands on an SSH server or client via an integer overflow.

SSH before 2.0, with RC4 encryption and the "disallow NULL passwords" option enabled, makes it easier for remote attackers to guess portions of user passwords by replaying user sessions with certain modifications, which trigger different messages depending on whether the guess is correct or not.

The SSH-1 protocol allows remote servers to conduct man-in-the-middle attacks and replay a client challenge response to a target server by creating a Session ID that matches the Session ID of the target, but which uses a public key pair that is weaker than the target's public key, which allows the attacker to compute the corresponding private key and use the target's Session ID with the compromised key pair to masquerade as the target.

The RC4 stream cipher as used by SSH1 allows remote attackers to modify messages without detection by XORing the original message's cyclic redundancy check (CRC) with the CRC of a mask consisting of all the bits of the original message that were modified.

The IDEA cipher as implemented by SSH1 does not protect the final block of a message against modification, which allows remote attackers to modify the block without detection by changing its cyclic redundancy check (CRC) to match the modifications to the message.

SSH before 2.0, when using RC4 and password authentication, allows remote attackers to replay messages until a new server key (VK) is generated.

SSH before 2.0 disables host key checking when connecting to the localhost, which allows remote attackers to silently redirect connections to the localhost by poisoning the client's DNS cache.

Directory traversal vulnerability in scp in sshd 1.2.xx allows a remote malicious scp server to overwrite arbitrary files via a .. (dot dot) attack.