CVE-2026-115913
Description détaillée
vLLM is an inference and serving engine for large language models. Prior to 0.32.0, the Phi-4 multimodal audio encoder in phi4mm_audio.py and phi4mm_utils.py uses eager attention by default, allowing a remote client to submit one legal long audio clip that causes MultiHeadedAttention.forward_attention to materialize and retain full attention matrices in self.attn. The scaled dot-product attention branch does not materialize these matrices. The resulting quadratic GPU-memory use can cause an out-of-memory condition and terminate EngineCore for all users. This issue is fixed in version 0.32.0.
Vecteur d'attaque (CVSS)
Dernières Vulnérabilités
CVE-2026-81509
nforwardauth is a forward auth service that allows the use of a single authentication middleware for multiple sites. Versions prior to 1.5.1 accept an unvalidated `r` query parameter on the `/login` endpoint and use them as the post-authentication redirect destination. An unauthenticated attacker can create a legitimate `nforwardauth` login URL that redirects a victim to an attacker-controlled website after the victim successfully authenticates. This issue may facilitate phishing and other social-engineering attacks by abusing the trust associated with the legitimate authentication domain. This vulnerability does not provide an unauthenticated ForwardAuth bypass. Version 1.5.1 contains a patch.
CVE-2026-72758
When parsing an AS range in Routinator, if the maximum value of the range is smaller than the minimum value of the range, Routinator will never reach the maximum value and run out of memory.
CVE-2026-72757
Routinator runs out of open connections on the HTTP server when many connections are opened but never closed, leaving the HTTP server unavailable. This condition can be triggered maliciously by an attacker by opening a large number of connections to the HTTP server. This only affects users that make their HTTP or RTR server available to untrusted networks.