Mbed TLS before 3.6.4 has a PEM parsing one-byte heap-based buffer underflow, in mbedtls_pem_read_buffer and two mbedtls_pk_parse functions, via untrusted PEM input.
Mbed TLS before 2.28.10 and 3.x before 3.6.3, on the client side, accepts servers that have trusted certificates for arbitrary hostnames unless the TLS client application calls mbedtls_ssl_set_hostname.
Mbed TLS before 2.28.10 and 3.x before 3.6.3, in some cases of failed memory allocation or hardware errors, uses uninitialized stack memory to compose the TLS Finished message, potentially leading to authentication bypasses such as replays.
An issue was discovered in Mbed TLS through 3.5.1. In mbedtls_ssl_session_reset, the maximum negotiable TLS version is mishandled. For example, if the last connection negotiated TLS 1.2, then 1.2 becomes the new maximum.
Use of a Broken or Risky Cryptographic Algorithm in the function mbedtls_mpi_exp_mod() in lignum.c in Mbed TLS Mbed TLS all versions before 3.0.0, 2.27.0 or 2.16.11 allows attackers with access to precise enough timing and memory access information (typically an untrusted operating system attacking a secure enclave such as SGX or the TrustZone secure world) to recover the private keys used in RSA.
In Mbed TLS before 3.1.0, psa_aead_generate_nonce allows policy bypass or oracle-based decryption when the output buffer is at memory locations accessible to an untrusted application.