vLLM is an inference and serving engine for large language models (LLMs). The SSRF protection fix for CVE-2026-24779 add in 0.15.1 can be bypassed in the load_from_url_async method due to inconsistent URL parsing behavior between the validation layer and the actual HTTP client. The SSRF fix uses urllib3.util.parse_url() to validate and extract the hostname from user-provided URLs. However, load_from_url_async uses aiohttp for making the actual HTTP requests, and aiohttp internally uses the yarl library for URL parsing. This vulnerability in 0.17.0.

vLLM is an inference and serving engine for large language models (LLMs). Starting in version 0.10.1 and prior to version 0.14.0, vLLM loads Hugging Face `auto_map` dynamic modules during model resolution without gating on `trust_remote_code`, allowing attacker-controlled Python code in a model repo/path to execute at server startup. An attacker who can influence the model repo/path (local directory or remote Hugging Face repo) can achieve arbitrary code execution on the vLLM host during model load. This happens before any request handling and does not require API access. Version 0.14.0 fixes the issue.

vllm-project vllm version 0.6.0 contains a vulnerability in the AsyncEngineRPCServer() RPC server entrypoints. The core functionality run_server_loop() calls the function _make_handler_coro(), which directly uses cloudpickle.loads() on received messages without any sanitization. This can result in remote code execution by deserializing malicious pickle data.

vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. Versions starting from 0.6.5 and prior to 0.8.5, having vLLM integration with mooncake, are vulnerable to remote code execution due to using pickle based serialization over unsecured ZeroMQ sockets. The vulnerable sockets were set to listen on all network interfaces, increasing the likelihood that an attacker is able to reach the vulnerable ZeroMQ sockets to carry out an attack. vLLM instances that do not make use of the mooncake integration are not vulnerable. This issue has been patched in version 0.8.5.

vLLM, an inference and serving engine for large language models (LLMs), has an issue in versions 0.6.5 through 0.8.4 that ONLY impacts environments using the `PyNcclPipe` KV cache transfer integration with the V0 engine. No other configurations are affected. vLLM supports the use of the `PyNcclPipe` class to establish a peer-to-peer communication domain for data transmission between distributed nodes. The GPU-side KV-Cache transmission is implemented through the `PyNcclCommunicator` class, while CPU-side control message passing is handled via the `send_obj` and `recv_obj` methods on the CPU side.​ The intention was that this interface should only be exposed to a private network using the IP address specified by the `--kv-ip` CLI parameter. The vLLM documentation covers how this must be limited to a secured network. The default and intentional behavior from PyTorch is that the `TCPStore` interface listens on ALL interfaces, regardless of what IP address is provided. The IP address given was only used as a client-side address to use. vLLM was fixed to use a workaround to force the `TCPStore` instance to bind its socket to a specified private interface. As of version 0.8.5, vLLM limits the `TCPStore` socket to the private interface as configured.

The October CMS debugbar plugin before version 3.1.0 contains a feature where it will log all requests (and all information pertaining to each request including session data) whenever it is enabled. This presents a problem if the plugin is ever enabled on a system that is open to untrusted users as the potential exists for them to use this feature to view all requests being made to the application and obtain sensitive information from those requests. There even exists the potential for account takeovers of authenticated users by non-authenticated public users, which would then lead to a number of other potential issues as an attacker could theoretically get full access to the system if the required conditions existed. Issue has been patched in v3.1.0 by locking down access to the debugbar to all users; it now requires an authenticated backend user with a specifically enabled permission before it is even usable, and the feature that allows access to stored request information is restricted behind a different permission that's more restrictive.

The CE21 Suite plugin for WordPress is vulnerable to Sensitive Information Exposure in all versions up to, and including, 2.3.1 via the log file. This makes it possible for unauthenticated attackers to extract sensitive data including authentication credentials, which can be used to log in as other users as long as they have used the plugin's custom authentication feature before. This may include administrators, which makes a complete site takeover possible.

API Connect V2018.1 through 2018.4.1.1 is impacted by access token leak. Authorization tokens in some URLs can result in the tokens being written to log files. IBM X-Force ID: 155626.

A Session ID leak in the DEBUG log file in Graylog before 4.1.2 allows attackers to escalate privileges (to the access level of the leaked session ID).

In the Rapid Gator application 0.7.1 for Android, the username and password are stored in the log during authentication, and may be available to attackers via logcat.

In the Orbitz application 19.31.1 for Android, the username and password are stored in the log during authentication, and may be available to attackers via logcat.

In the DoorDash application through 11.5.2 for Android, the username and password are stored in the log during authentication, and may be available to attackers via logcat.

In the Seesaw Parent and Family application 6.2.5 for Android, the username and password are stored in the log during authentication, and may be available to attackers via logcat.

In the PowerSchool Mobile application 1.1.8 for Android, the username and password are stored in the log during authentication, and may be available to attackers via logcat.

In the Dark Horse Comics application 1.3.21 for Android, token information (equivalent to the username and password) is stored in the log during authentication, and may be available to attackers via logcat.

Atlantis is a self-hosted golang application that listens for Terraform pull request events via webhooks. Atlantis logs contains GitHub credentials (tokens `ghs_...`) when they are rotated. This enables an attacker able to read these logs to impersonate Atlantis application and to perform actions on GitHub. When Atlantis is used to administer a GitHub organization, this enables getting administration privileges on the organization. This was reported in #4060 and fixed in #4667 . The fix was included in Atlantis v0.30.0. All users are advised to upgrade. There are no known workarounds for this vulnerability.

A flaw was found in, all under 2.0.20, in the Undertow DEBUG log for io.undertow.request.security. If enabled, an attacker could abuse this flaw to obtain the user's credentials from the log files.

A Session ID leak in the audit log in Graylog before 4.1.2 allows attackers to escalate privileges (to the access level of the leaked session ID).

JunoClaw is an agentic AI platform built on Juno Network. Prior to 0.x.y-security-1, every MCP write tool (send_tokens, execute_contract, instantiate_contract, upload_wasm, ibc_transfer, etc.) accepted 'mnemonic: string' as an explicit tool-call parameter. The BIP-39 seed was consequently embedded in the LLM tool-call JSON, exposing it to any transport, log, or telemetry surface in the path between the LLM provider and the MCP process. This vulnerability is fixed in 0.x.y-security-1.

A sensitive data disclosure flaw was found in the way Logstash versions before 5.6.15 and 6.6.1 logs malformed URLs. If a malformed URL is specified as part of the Logstash configuration, the credentials for the URL could be inadvertently logged as part of the error message.

Valtimo is an open source business process and case management platform. When opening a form in Valtimo, the access token (JWT) of the user is exposed to `api.form.io` via the the `x-jwt-token` header. An attacker can retrieve personal information from this token, or use it to execute requests to the Valtimo REST API on behalf of the logged-in user. This issue is caused by a misconfiguration of the Form.io component. The following conditions have to be met in order to perform this attack: An attacker needs to have access to the network traffic on the `api.form.io` domain; the content of the `x-jwt-token` header is logged or otherwise available to the attacker; an attacker needs to have network access to the Valtimo API; and an attacker needs to act within the time-to-live of the access token. The default TTL in Keycloak is 5 minutes. Versions 10.8.4, 11.1.6 and 11.2.2 have been patched.

A vulnerability was found in Undertow web server before 2.0.21. An information exposure of plain text credentials through log files because Connectors.executeRootHandler:402 logs the HttpServerExchange object at ERROR level using UndertowLogger.REQUEST_LOGGER.undertowRequestFailed(t, exchange)

Cron log backup files contain administrator session IDs. It is trivial for any attacker who can reach the Pandora FMS Console to scrape the cron logs directory for cron log backups. The contents of these log files can then be abused to authenticate to the application as an administrator. This issue affects Pandora FMS <= 772.

A flaw was found in the way Ansible (2.3.x before 2.3.3, and 2.4.x before 2.4.1) passed certain parameters to the jenkins_plugin module. Remote attackers could use this flaw to expose sensitive information from a remote host's logs. This flaw was fixed by not allowing passwords to be specified in the "params" argument, and noting this in the module documentation.