The flaw is an out‑of‑bounds read in mbedtls_ccm_finish() that occurs when the multipart CCM API is called with a tag_len value larger than the 16‑byte authentication buffer. The read exposes adjacent memory belonging to the CCM context, allowing an attacker to retrieve sensitive data such as cryptographic keys or authentication tags. This vulnerability is classified as CWE‑125 and results in potential information disclosure.

The vulnerability affects the Arm mbed TLS library. Public versions in the 3.x series prior to 3.6.6 are susceptible. Versions of Mbed TLS 4.x before the fix also contain the same internal validation error, but the API is not publicly exposed. Applications that use the multipart CCM API in these versions are at risk.

The CVSS score of 7.5 indicates a high severity, while an EPSS of less than 1% points to a low likelihood of exploitation in the wild. The vulnerability is not listed in the CISA KEV catalog. Exploitation requires that an attacker can invoke the multipart CCM API with an oversized tag_len, which is typically possible only through application‑level code that accepts user‑supplied parameters. The attack vector is inferred to be application‑level and could occur remotely if the application processes external input. The patch in 3.6.6 eliminates the unvalidated length check, mitigating the read.