CVE-2026-107858
HighCVSS 7.2Exploitation Probability (EPSS)
Low risk30th percentile - higher than 30% of all known CVEs
Summary
The development web server (ADK Web) in Google Agent Development Kit (ADK) for Java lacks WebSocket origin validation and uses a permissive cross-domain policy. This allows a remote attacker to control a locally running agent and read its data via a malicious web page visited by a developer while the server is running. The issue is fixed in version 1.13.0.
Risk Assessment
A developer visiting a malicious page while the server runs may trigger agent tools, leak data, and list, read, create, or delete sessions and artifacts. Lack of authentication and localhost binding do not prevent exploitation.
Recommendation
Update ADK to version 1.13.0. In versions 1.6.0 through 1.12.0, set the adk.web.cors.origins property to an explicit allowlist of trusted origins.
Related vulnerabilities
- CVE-2026-121953Critical
FreeRDP before 3.32.2 has a use-after-free in the client's rdpgfx dynamic virtual channel: a duplicate-name channel with a new ChannelId overwrites the listener's single callback slot, which rdpgfx_on_close frees without clearing. A malicious RDP server can open and close such a channel, then send a PDU on the original channel to crash the client or potentially execute code.
- CVE-2026-121952Critical
FreeRDP before 3.32.2 has an out-of-bounds write in the proxy because rdp_read_bitmap_cache_v2_capability_set rejects only numCellCaches values above 5, yet the bitmap cache v2 cell array is later accessed at fixed indices 0 through 4. A client connecting to the proxy can declare fewer than five cell caches in its Confirm Active, causing heap corruption, heap data leakage to the backend, or a crash.
- CVE-2026-122006Critical
The activestorage gem before 8.1.4.1 for Ruby on Rails, when ImageMagick attachment processing occurs, can allow server read and write operations (e.g., reading secret_key_base) because the ImageMagick analyzer runs when a file is attached, and there is not an ImageMagick security policy to block unintended functionality.
- CVE-2026-121640Critical
An Authentication Bypass by Capture-replay vulnerability in WatchGuard AuthPoint Gateway allows an attacker who has observed a legitimate request between AuthPoint Gateway components to replay it and perform unauthorized administrative actions, including disrupting authentication services.
- CVE-2026-121639Critical
A Missing Authentication for Critical Function vulnerability in WatchGuard AuthPoint Gateway allows a remote unauthenticated attacker with network access to the AuthPoint Gateway host to disrupt authentication services and delete the local directory synchronization configuration.
- CVE-2026-121638Critical
An Improper Certificate Validation vulnerability in WatchGuard AuthPoint Gateway allows an attacker with an on-path network position to intercept and modify communications between the AuthPoint Gateway and local authentication providers.
- CVE-2026-65555Critical
The Uncanny Toolkit for LearnDash plugin (through version 3.8.2) contains an incorrect privilege assignment vulnerability that allows privilege escalation. A low-privileged user can gain access to functionality reserved for higher roles.
- CVE-2026-119481Critical
The JetFormBuilder plugin (through version 3.6.7.1) contains an incorrect privilege assignment vulnerability enabling privilege escalation. A low-privileged user can perform actions reserved for higher roles.
- CVE-2026-114425Critical
SEPPmail Secure Email Gateway before 15.1.1 trusts a file path supplied in externally provided Large File Transfer attachment metadata. An unauthenticated remote sender can cause the mail-processing worker to disclose, move, or delete an accessible regular file when LFT is enabled and the message reaches an affected secure-mail processing path. The validated workers ran as root.
- CVE-2026-114424Critical
SEPPmail Secure Email Gateway before 15.1.1 can evaluate attacker-influenced replacement text as Perl code in sender or recipient replacement rules. An unauthenticated remote sender can execute operating-system commands in the mail-processing worker context when an active rule uses message-derived replacement data and its search condition matches. The validated worker context was root.
Original NVD description (English source)
Missing Origin Validation in WebSockets (CWE-1385) and Permissive Cross-domain Policy with Untrusted Domains (CWE-942) in the development web server (ADK Web) in Google Agent Development Kit (ADK) for Java (com.google.adk:google-adk-dev versions 0.1.0 before 1.13.0 and com.google.adk:google-adk-maven-plugin versions 0.3.0 before 1.13.0) allow a remote attacker to control a locally running agent and read its data via a malicious web page visited by a developer while the server is running. The /run_live WebSocket endpoint accepts connections from any origin by default (hardcoded via setAllowedOrigins("*") in versions 0.1.0 through 1.5.0, and defaulting to * via adk.web.cors.origins in versions 1.6.0 through 1.12.0), and since version 0.4.0 the REST API returns Access-Control-Allow-Origin: * by default. Because the server has no authentication and requests originate from the developer's browser (so binding to localhost does not prevent exploitation), a cross-origin page can send messages to the agent (triggering registered tools), stream agent responses, and list, read, create, or delete sessions and artifacts. In versions 1.6.0 through 1.12.0, users can mitigate the issue by setting the adk.web.cors.origins property to an explicit allowlist of trusted origins (such as http://localhost:8000). The vulnerability is fixed in version 1.13.0, which enforces a same-origin default for CORS and the /run_live WebSocket endpoint, and rejects DNS-rebinding requests over loopback.
Vulnerability data from NVD (NIST) · CISA KEV · EPSS

