CVE Watch14 de out. de 2026

CVE-2026-107292

Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. From 1.34.0 until 2.30.0, the Agent.to_w

CVSS

6.4

Média

EPSS

0.1%

KEV

—

Exploit Today

0-100

Publicado: 8 de out. de 2026 · Última mod.: 14 de out. de 2026 · CWE-346 · CWE-350

EPSS · 30d

0.1%EPSS · 30 dias0.1%

2026-10-092026-11-05

Descrição técnica

Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. From 1.34.0 until 2.30.0, the Agent.to_web() and clai web development chat server does not validate the Host header, allowing a website visited by a developer to use DNS rebinding to reach a loopback-hosted agent as a same-origin service. The hostile page can read the served UI and submit chat requests that execute agent tools with the local process's privileges and credentials, causing data disclosure or unwanted side effects. Binding to localhost, Origin checks, and CSRF tokens do not prevent the same-origin DNS rebinding path. This issue is fixed in versions 1.107.5 and 2.30.0.

Referências oficiais

CVEs relacionados

CVECVSSEPSSKEVRExploitTítuloMod.

CVE-2026-1232728.0 ALT

—

——0Mark3labs mcp-go before v1.2.1 contains an improper validation vulnerability in OAuthHandler.fetchServerMetadata that allows malicious MCP servers to obtain tokens for other resources by declaring arbitrary resources in well-known metadata. Attackers can declare a different MCP server as the RFC 8707 resource indicator, causing the authorization server to mint tokens the attacker replays against that resource.16h CVE-2026-1232718.0 ALT

—

——0ChatCLI before v1.215.2 contains an improper validation vulnerability in the MCP OAuth login flow that uses server-supplied Protected Resource Metadata as the resource indicator without checking it matches the configured server URL. A malicious MCP server can declare another resource on the same authorization server, obtain the user's bearer token for it, and replay that token against the other resource.18h CVE-2026-119344—

—

——0Cocos AI is a confidential computing system for AI. In version 0.8.2 and earlier, Cocos AI accepts trusted execution environment attestation Evidence when firmware patch levels and software measurements match, without binding the Evidence to an allowed platform owner or unique machine identity and without enforcing revocation. A remote attacker who controls or compromises another platform trusted by the same verifier can substitute valid Evidence from that platform and cause authorization of the wrong platform. This vulnerability is addressed in 0.9.0. However, this vulnerability can't be patched in source code alone, since the vulnerability arises from the confidential computing threat model itself.21h CVE-2026-1028959.8 CRÍ

26.4%

——8GPUStack is an open-source GPU cluster manager for AI model serving and GPU instance provisioning. Prior to 2.2.0, get_current_user in gpustack/api/auth.py treated a loopback request source as the built-in administrator when force_auth_localhost was false, which was the default. A remote request could appear to originate from 127.0.0.1 because client_ip_getter accepted a same-host reverse proxy peer or a client-supplied X-GPUStack-Real-IP loopback value. The bypass allowed unauthenticated administrative access, including user and model changes, persistent administrator API-key creation, disclosure of worker and cloud-provider secrets, and code execution through POST /v2/inference-backends/from-yaml. Exploitation was observed in the wild against a 2.1.1 deployment. This issue is fixed in version 2.2.0.8h CVE-2026-63183—

7.7%

——2OmniAuth is a flexible authentication system utilizing Rack middleware. Prior to 2.2.0, OmniAuth derives provider callback URLs from request-controlled host information when neither a fixed full_host nor framework-level incoming host validation is configured. A remote unauthenticated client can supply a forged Host header, creating a host header injection risk in applications whose web framework does not verify incoming host headers. Applications using appropriately configured Rails host authorization or a fixed full_host are not affected by this path. This issue is fixed in version 2.2.0.17h CVE-2026-927048.3 ALT

5.0%

——2SharpEmu is an experimental PlayStation 5 emulator for Windows and Linux. In version 0.0.3-release.3 and earlier, the repository-provided debugger frontend tooling is vulnerable while running. This tooling is not included in the release archives. The frontend exposes an HTTP API on 127.0.0.1:8765 without authentication or CSRF token. Its do_POST handler in tools/SharpEmu.DebuggerFrontend/debugger_frontend.py does not validate the Origin, Referer, Sec-Fetch-Site, or Host headers. The _read_json_body() function also accepts JSON sent with browser-simple content types instead of requiring application/json . A malicious page that can reach the loopback listener can send cross-origin requests to /api/launch , /api/command , and /api/connect . Chromium may require Local Network Access permission for pages served from public origins. Local-origin pages and other delivery paths are not consistently subject to this restriction. The /api/launch endpoint accepts an attacker-selected emulatorPath if it points to an existing file on the victim’s host. On POSIX systems, that file must also be executable. The endpoint additionally requires an existing ebootPath . The /api/command endpoint exposes debugger operations, including write-memory , set-register , and continue . The /api/connect endpoint can initiate outbound TCP connections. These operations can let an attacker execute an existing binary with the victim’s privileges, control the attached emulated process, or reach network services accessible from the host. DNS rebinding can additionally let the attacker read API responses interactively. This issue is fixed in version 0.0.3-release.3-patch.12d