Out-of-bounds bad_words token indices in vLLM through 0.29.0 are not validated against the model's generation output width, allowing the indices to be written past the logits buffer and corrupt the sampling state of concurrently batched requests, so unrelated in-flight HTTP requests can receive incorrect output tokens. Exploitation requires API access sufficient to submit sampling requests with a bad_words parameter (authenticated, PR:L), a target model whose tokenizer exposes input-only placeholder tokens with IDs above the model's logits vocabulary (typically multimodal models such as those with <|audio|> tokens), and concurrent requests sharing the batch, so a text-only model whose tokenizer vocabulary equals its output vocabulary is not affected. No public exploit code has been identified at time of analysis and there is no KEV listing; the issue is a low-severity integrity/quality-of-output defect rather than a confidentiality or availability compromise, and an upstream fix is available as a pull request.
Path traversal in rclone before 1.75.1 allows object names returned in server or third-party directory-listing responses to contain forward slashes and '../' parent-directory references that are not confined to the listed directory, so a malicious or compromised source can attempt to steer writes outside the transfer destination root. Exploitation requires the victim to run rclone against an attacker-influenced remote (attacker-controlled source), user interaction to initiate the transfer, and high attack complexity; the local backend's existing downstream protections currently block actual file escape on the default local backend, keeping real-world impact limited. Rated only 2.3 (CVSS 4.0) and assessed as CVSS:3.1/AV:N/AC:H/PR:N/UI:R/S:U/C:N/I:L/A:N, this is a low-priority defense-in-depth hardening issue rather than an actionable threat, and no public exploit code was identified at time of analysis (no CISA KEV listing).
Cross-site scripting in Hono before 4.13.7 allows unauthenticated remote attackers to inject arbitrary HTML and execute script in a victim's browser under the application's origin when the application uses hono/jsx server-side rendering in one of four specific patterns involving plain strings: as a child or fallback of Suspense, as a string child of ErrorBoundary alongside an asynchronous sibling, as the single child of Context.Provider, or as the root value passed to renderToString() or renderToReadableStream() from hono/jsx/dom/server. Exploitation requires attacker-controlled data to flow directly into those strings during SSR and a victim to load the rendered output, making this a low-priority XSS despite the network attack vector and no authentication requirement. No public exploit code has been identified at time of analysis, and the issue is fixed in Hono 4.13.7.