An issue was discovered in Arm Mbed TLS before 2.16.6 and 2.7.x before 2.7.15. An attacker that can get precise enough side-channel measurements can recover the long-term ECDSA private key by (1) reconstructing the projective coordinate of the result of scalar multiplication by exploiting side channels in the conversion to affine coordinates; (2) using an attack described by Naccache, Smart, and Stern in 2003 to recover a few bits of the ephemeral scalar from those projective coordinates via several measurements; and (3) using a lattice attack to get from there to the long-term ECDSA private key used for the signatures. Typically an attacker would have sufficient access when attacking an SGX enclave and controlling the untrusted OS.

An issue was discovered in Mbed TLS before 2.28.2 and 3.x before 3.3.0. An adversary with access to precise enough information about memory accesses (typically, an untrusted operating system attacking a secure enclave) can recover an RSA private key after observing the victim performing a single private-key operation, if the window size (MBEDTLS_MPI_WINDOW_SIZE) used for the exponentiation is 3 or smaller.

Arm Mbed TLS before 2.14.1, before 2.7.8, and before 2.1.17 allows a local unprivileged attacker to recover the plaintext of RSA decryption, which is used in RSA-without-(EC)DH(E) cipher suites.

Arm Mbed TLS before 2.19.0 and Arm Mbed Crypto before 2.0.0, when deterministic ECDSA is enabled, use an RNG with insufficient entropy for blinding, which might allow an attacker to recover a private key via side-channel attacks if a victim signs the same message many times. (For Mbed TLS, the fix is also available in versions 2.7.12 and 2.16.3.)

ARM mbed TLS before 2.1.11, before 2.7.2, and before 2.8.0 has a buffer over-read in ssl_parse_server_psk_hint() that could cause a crash on invalid input.

An issue was discovered in Mbed TLS versions from 2.19.0 up to 3.6.5, Mbed TLS 4.0.0. Insufficient protection of serialized SSL context or session structures allows an attacker who can modify the serialized structures to induce memory corruption, leading to arbitrary code execution. This is caused by Incorrect Use of Privileged APIs.

ARM mbed TLS before 2.1.11, before 2.7.2, and before 2.8.0 has a buffer over-read in ssl_parse_server_key_exchange() that could cause a crash on invalid 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.

An issue was discovered in Mbed TLS before 2.28.2 and 3.x before 3.3.0. There is a potential heap-based buffer overflow and heap-based buffer over-read in DTLS if MBEDTLS_SSL_DTLS_CONNECTION_ID is enabled and MBEDTLS_SSL_CID_IN_LEN_MAX > 2 * MBEDTLS_SSL_CID_OUT_LEN_MAX.

Mbed TLS before 3.0.1 has a double free in certain out-of-memory conditions, as demonstrated by an mbedtls_ssl_set_session() failure.

An issue was discovered in Mbed TLS before 2.28.1 and 3.x before 3.2.0. In some configurations, an unauthenticated attacker can send an invalid ClientHello message to a DTLS server that causes a heap-based buffer over-read of up to 255 bytes. This can cause a server crash or possibly information disclosure based on error responses. Affected configurations have MBEDTLS_SSL_DTLS_CLIENT_PORT_REUSE enabled and MBEDTLS_SSL_IN_CONTENT_LEN less than a threshold that depends on the configuration: 258 bytes if using mbedtls_ssl_cookie_check, and possibly up to 571 bytes with a custom cookie check function.

ARM mbed TLS before 1.3.21 and 2.x before 2.1.9, if optional authentication is configured, allows remote attackers to bypass peer authentication via an X.509 certificate chain with many intermediates. NOTE: although mbed TLS was formerly known as PolarSSL, the releases shipped with the PolarSSL name are not affected.

Integer Overflow vulnerability in Mbed TLS 2.x before 2.28.7 and 3.x before 3.5.2, allows attackers to cause a denial of service (DoS) via mbedtls_x509_set_extension().

Mbed TLS 2.x before 2.28.5 and 3.x before 3.5.0 has a Buffer Overflow.

A vulnerability has been identified in APOGEE PXC Compact (BACnet) (All versions < V3.5.5), APOGEE PXC Compact (P2 Ethernet) (All versions < V2.8.20), APOGEE PXC Modular (BACnet) (All versions < V3.5.5), APOGEE PXC Modular (P2 Ethernet) (All versions < V2.8.20), Nucleus NET (All versions < V5.2), Nucleus ReadyStart V3 (All versions < V2012.12), Nucleus Source Code (All versions), PLUSCONTROL 1st Gen (All versions), TALON TC Compact (BACnet) (All versions < V3.5.5), TALON TC Modular (BACnet) (All versions < V3.5.5). Initial Sequence Numbers (ISNs) for TCP connections are derived from an insufficiently random source. As a result, the ISN of current and future TCP connections could be predictable. An attacker could hijack existing sessions or spoof future ones.

hw: cpu: speculative store bypass

Systems with microprocessors utilizing speculative execution and speculative execution of memory reads before the addresses of all prior memory writes are known may allow unauthorized disclosure of information to an attacker with local user access via a side-channel analysis, aka Speculative Store Bypass (SSB), Variant 4.

hw: cpu: speculative execution bounds-check bypass

Systems with microprocessors utilizing speculative execution and branch prediction may allow unauthorized disclosure of information to an attacker with local user access via a side-channel analysis.

hw: cpu: speculative execution permission faults handling

Systems with microprocessors utilizing speculative execution and indirect branch prediction may allow unauthorized disclosure of information to an attacker with local user access via a side-channel analysis of the data cache.

In Arm ArmNN through 2026-03-27, an integer overflow in TensorShape::GetNumElements() in armnn/Tensor.cpp allows a crafted TFLite model file to bypass buffer size validation and trigger a heap-based buffer over-read during model optimization. The overflow occurs when multiplying tensor dimensions using 32-bit unsigned arithmetic without overflow detection, causing GetNumBytes() to return an understated allocation size. During Optimize()->InferOutputShapes(), the BatchToSpaceNdLayer reads beyond the allocated buffer.

Potential SME Memory Access Issue due to TLBI+DSB Failure on Arm C1-Pro

An issue has been identified in Arm C1-Pro before r1p2-50eac0, where, under certain conditions, a TLBI+DSB might fail to ensure the completion of memory accesses related to SME.