Vulnerabilidades do Red Hat Openshift Ai

Traefik before 2.10.5 and 3.0.0-beta4 is affected by a denial-of-service vulnerability in HTTP/2 request handling inherited from the Go standard library's HTTP/2 implementation (CVE-2023-44487 / CVE-2023-39325, the 'Rapid Reset' technique). A remote attacker can rapidly create and cancel HTTP/2 streams to exhaust server resources and cause service unavailability.

JavaScript Cookie is a JavaScript API for handling cookies, client-side. Prior to version 3.0.7, js-cookie's internal assign() helper copies properties with for...in + plain assignment. When the source object is produced by JSON.parse, the JSON object's "__proto__" member is an own enumerable property, so the for…in enumerates it and the target[key] = source[key] write triggers the Object.prototype.__proto__ setter on the fresh target ({}). The result is a per-instance prototype hijack: Object.prototype itself is untouched, but the merged attributes object now inherits attacker-controlled keys. Because the consuming set() function then enumerates the merged object with another for...in, every key the attacker placed on the polluted prototype lands in the resulting Set-Cookie string as an attribute pair. The attacker can set domain=, secure=, samesite=, expires=, and path= on cookies whose attributes the developer thought were locked down. This issue has been patched in version 3.0.7.

LiteLLM is a proxy server (AI Gateway) to call LLM APIs in OpenAI (or native) format. From version 1.74.2 to before version 1.83.7, two endpoints used to preview an MCP server before saving it — POST /mcp-rest/test/connection and POST /mcp-rest/test/tools/list — accepted a full server configuration in the request body, including the command, args, and env fields used by the stdio transport. When called with a stdio configuration, the endpoints attempted to connect, which spawned the supplied command as a subprocess on the proxy host with the privileges of the proxy process. The endpoints were gated only by a valid proxy API key, with no role check. Any authenticated user — including holders of low-privilege internal-user keys — could therefore run arbitrary commands on the host. This issue has been patched in version 1.83.7.

A vulnerability in the `TFSMLayer` class of the `keras` package, version 3.13.0, allows attacker-controlled TensorFlow SavedModels to be loaded during deserialization of `.keras` models, even when `safe_mode=True`. This bypasses the security guarantees of `safe_mode` and enables arbitrary attacker-controlled code execution during model inference under the victim's privileges. The issue arises due to the unconditional loading of external SavedModels, serialization of attacker-controlled file paths, and the lack of validation in the `from_config()` method.

A flaw was found in Red Hat OpenShift AI (RHOAI) llama-stack-operator. This vulnerability allows unauthorized access to Llama Stack services deployed in other namespaces via direct network requests, because no NetworkPolicy restricts access to the llama-stack service endpoint. As a result, a user in one namespace can access another user’s Llama Stack instance and potentially view or manipulate sensitive data.