FFmpeg CVE-2026-107660

Exploitation requires ALL of the following specific conditions: (1) FFmpeg built against the mbedTLS TLS backend (the vulnerable tls_open() lives in libavformat/tls_mbedtls.c, not the OpenSSL/GnuTLS backends); (2) the client connecting to an IP-literal URL (e.g. https://198.51.100.5/) over https, rtmps, or tls - connections to DNS hostnames are NOT affected because hostname verification still occurs for them; and (3) an attacker occupying a man-in-the-middle position on the network path who possesses any certificate signed by a trusted CA. Limiting factors that reduce realistic risk: the High attack complexity and Present attack requirement (AT:P) mean the MITM position must already be established, the non-default mbedTLS backend narrows the affected population, and only IP-literal target URLs are vulnerable; impact is confined to reading and tampering with the stream (confidentiality/integrity), with no availability or code-execution effect.

This is a genuine but moderate-priority issue, best characterized as a high-impact-in-narrow-conditions flaw rather than a broadly dangerous one. The CVSS 4.0 vector (AV:N/AC:H/AT:P/PR:N/UI:N) tells the real story: although network-reachable and requiring no privileges or user interaction, it carries High attack complexity and a Present attack requirement (AT:P), meaning the attacker must already hold a privileged man-in-the-middle network position and the victim must be connecting to an IP-literal URL. Impact is limited (VC:L/VI:L/VA:N) - confidentiality and integrity of the stream, no availability loss - consistent with a TLS interception/tampering bug rather than code execution. No EPSS score, CISA KEV listing, or public POC was provided, so there is no public exploit identified at time of analysis and no evidence of active exploitation; the AT:P/AC:H profile also argues against opportunistic mass exploitation and toward targeted, position-dependent attacks. The signals are internally consistent (CWE-297 host-mismatch matches the described IP-literal hostname-check skip, and the 6.3/Medium score fits the limited-impact MITM nature). One caveat defenders should verify: exploitation is specific to FFmpeg builds using the mbedTLS backend (tls_mbedtls.c), which is not FFmpeg's most common TLS backend, further narrowing the real-world exposed population below what a raw 'FFmpeg' product match would suggest.

An attacker positioned on the network path (e.g., rogue Wi-Fi, ARP/DNS spoofing, or a compromised upstream hop) intercepts an FFmpeg client's https/rtmps/tls connection to an IP-literal URL such as https://203.0.113.10/stream. Because tls_open() in the mbedTLS backend skips hostname verification for IP-address hosts, the attacker presents any certificate validly signed by a trusted CA and the connection succeeds, letting them read and modify the media/data stream in transit. No public exploit code was identified in the provided data, and the attack requires both a pre-existing MITM position and the victim targeting an IP literal rather than a DNS name.