Latest Vulnerabilities with Public Exploits

Cursor is a code editor built for programming with AI. Sandbox escape via writing .git configuration was possible in versions prior to 2.5. A malicious agent (ie prompt injection) could write to improperly protected .git settings, including git hooks, which may cause out-of-sandbox RCE next time they are triggered. No user interaction was required as Git executes these commands automatically. Fixed in version 2.5.

In OpenBSD through 7.8, the slaacd and rad daemons have an infinite loop when they receive a crafted ICMPv6 Neighbor Discovery (ND) option (over a local network) with length zero, because of an "nd_opt_len * 8 - 2" expression with no preceding check for whether nd_opt_len is zero.

Deserialization of untrusted data in Microsoft Office SharePoint allows an authorized attacker to execute code over a network.

An issue was discovered in 6.0 before 6.0.4, 5.2 before 5.2.13, and 4.2 before 4.2.30. `MultiPartParser` allows remote attackers to degrade performance by submitting multipart uploads with `Content-Transfer-Encoding: base64` including excessive whitespace. Earlier, unsupported Django series (such as 5.0.x, 4.1.x, and 3.2.x) were not evaluated and may also be affected. Django would like to thank Seokchan Yoon for reporting this issue.

A flaw has been found in Ollama up to 18.1. This issue affects some unknown processing of the file server/download.go of the component Model Pull API. Executing a manipulation can lead to server-side request forgery. The attack can be launched remotely. The vendor was contacted early about this disclosure but did not respond in any way.

exiftool-vendored provides cross-platform Node.js access to ExifTool. Prior to 35.19.0, exiftool-vendored starts ExifTool in -stay_open True -@ - mode, where arguments are read from stdin one per line. In affected versions, several caller-supplied strings were interpolated into ExifTool arguments without rejecting line delimiters. A newline or carriage return inside one of those strings could split a single intended argument into multiple ExifTool arguments, allowing argument injection. The fix also rejects NUL bytes as unsafe control characters. Applications that pass attacker-controlled strings to affected APIs may allow an attacker to make ExifTool read files accessible to the ExifTool process, or write output to attacker-chosen file system paths accessible to that process. No remote code execution has been demonstrated. This vulnerability is fixed in 35.19.0.

A flaw has been found in code-projects Visitor Management System 1.0. Affected by this issue is some unknown functionality of the file /vms/php/phone_0.php. This manipulation of the argument phone causes sql injection. The attack may be initiated remotely. The exploit has been published and may be used.

Reliance on untrusted inputs in a security decision in Microsoft Office Word allows an unauthorized attacker to bypass a security feature locally.

Untrusted pointer dereference in Windows HTTP.sys allows an authorized attacker to elevate privileges locally.

Improper Control of Generation of Code ('Code Injection') vulnerability in Funnelforms LLC FunnelFormsPro allows Remote Code Inclusion.This issue affects FunnelFormsPro: from n/a through 3.8.1.

In MariaDB server version through 11.8.5, when server audit plugin is enabled with server_audit_events variable configured with QUERY_DCL, QUERY_DDL, or QUERY_DML filtering, if an authenticated database user invokes a SQL statement prefixed with double-hyphen (—) or hash (#) style comments, the statement is not logged.

A vulnerability was detected in sgl-project SGLang up to 0.5.9. Impacted is the function get_tokenizer of the file python/sglang/srt/utils/hf_transformers_utils.py of the component HuggingFace Transformer Handler. The manipulation of the argument trust_remote_code with the input False as part of Boolean results in code injection. The attack can be executed remotely. A high complexity level is associated with this attack. The exploitability is considered difficult. In get_tokenizer(), when the caller passes trust_remote_code=False and HuggingFace transformers v5 returns a TokenizersBackend instance (the generic fallback for tokenizer classes not in the registry), SGLang silently re-invokes AutoTokenizer.from_pretrained with trust_remote_code=True, overriding the caller's explicit security setting. A model repository containing a malicious tokenizer.py referenced via auto_map in tokenizer_config.json will execute arbitrary Python in the SGLang process during this second call. No log line or warning is emitted. The override affects all current SGLang versions because transformers==5.3.0 is pinned in pyproject.toml. Both tokenizer_mode="auto" and tokenizer_mode="slow" are affected. The exploit is now public and may be used. The vendor was contacted early about this disclosure but did not respond in any way.

In the Linux kernel, the following vulnerability has been resolved: netfilter: nf_tables: fix use-after-free in nf_tables_addchain() nf_tables_addchain() publishes the chain to table->chains via list_add_tail_rcu() (in nft_chain_add()) before registering hooks. If nf_tables_register_hook() then fails, the error path calls nft_chain_del() (list_del_rcu()) followed by nf_tables_chain_destroy() with no RCU grace period in between. This creates two use-after-free conditions: 1) Control-plane: nf_tables_dump_chains() traverses table->chains under rcu_read_lock(). A concurrent dump can still be walking the chain when the error path frees it. 2) Packet path: for NFPROTO_INET, nf_register_net_hook() briefly installs the IPv4 hook before IPv6 registration fails. Packets entering nft_do_chain() via the transient IPv4 hook can still be dereferencing chain->blob_gen_X when the error path frees the chain. Add synchronize_rcu() between nft_chain_del() and the chain destroy so that all RCU readers -- both dump threads and in-flight packet evaluation -- have finished before the chain is freed.

vLLM is an inference and serving engine for large language models (LLMs). From 0.16.0 to before 0.19.0, a server-side request forgery (SSRF) vulnerability in download_bytes_from_url allows any actor who can control batch input JSON to make the vLLM batch runner issue arbitrary HTTP/HTTPS requests from the server, without any URL validation or domain restrictions. This can be used to target internal services (e.g. cloud metadata endpoints or internal HTTP APIs) reachable from the vLLM host. This vulnerability is fixed in 0.19.0.

The frontend gRPC server's streaming interceptor chain did not include the authorization interceptor. When a ClaimMapper and Authorizer are configured, unary RPCs enforce authentication and authorization, but the streaming AdminService/StreamWorkflowReplicationMessages endpoint accepted requests without credentials. This endpoint is registered on the same port as WorkflowService and cannot be disabled independently. An attacker with network access to the frontend port could open the replication stream without authentication. Data exfiltration is possible, but  only when a configured replication target is correctly configured and the attacker has knowledge of the cluster configuration, as the history service validates cluster IDs and peer membership before returning replication data. Temporal Cloud is not affected.

Rack::Session is a session management implementation for Rack. From 2.0.0 to before 2.1.2, Rack::Session::Cookie incorrectly handles decryption failures when configured with secrets:. If cookie decryption fails, the implementation falls back to a default decoder instead of rejecting the cookie. This allows an unauthenticated attacker to supply a crafted session cookie that is accepted as valid session data without knowledge of any configured secret. Because this mechanism is used to load session state, an attacker can manipulate session contents and potentially gain unauthorized access. This vulnerability is fixed in 2.1.2.

The Angular SSR is a server-rise rendering tool for Angular applications. Versions prior to 21.2.0-rc.1, 21.1.5, 20.3.17, and 19.2.21 have a Server-Side Request Forgery (SSRF) vulnerability in the Angular SSR request handling pipeline. The vulnerability exists because Angular’s internal URL reconstruction logic directly trusts and consumes user-controlled HTTP headers specifically the Host and `X-Forwarded-*` family to determine the application's base origin without any validation of the destination domain. Specifically, the framework didn't have checks for the host domain, path and character sanitization, and port validation. This vulnerability manifests in two primary ways: implicit relative URL resolution and explicit manual construction. When successfully exploited, this vulnerability allows for arbitrary internal request steering. This can lead to credential exfiltration, internal network probing, and a confidentiality breach. In order to be vulnerable, the victim application must use Angular SSR (Server-Side Rendering), the application must perform `HttpClient` requests using relative URLs OR manually construct URLs using the unvalidated `Host` / `X-Forwarded-*` headers using the `REQUEST` object, the application server must be reachable by an attacker who can influence these headers without strict validation from a front-facing proxy, and the infrastructure (Cloud, CDN, or Load Balancer) must not sanitize or validate incoming headers. Versions 21.2.0-rc.1, 21.1.5, 20.3.17, and 19.2.21 contain a patch. Some workarounds are available. Avoid using `req.headers` for URL construction. Instead, use trusted variables for base API paths. Those who cannot upgrade immediately should implement a middleware in their `server.ts` to enforce numeric ports and validated hostnames.

A vulnerability has been found in CodeWise Tornet Scooter Mobile App 4.75 on iOS/Android. The impacted element is an unknown function of the file /TwoFactor. Such manipulation leads to improper restriction of excessive authentication attempts. The attack may be performed from remote. Attacks of this nature are highly complex. The exploitability is regarded as difficult. The exploit has been disclosed to the public and may be used. The vendor was contacted early about this disclosure but did not respond in any way.

Improper Input Validation, Improper Control of Generation of Code ('Code Injection') vulnerability in Apache ActiveMQ, Apache ActiveMQ Broker, Apache ActiveMQ All. An authenticated attacker can use the admin web console page to construct a malicious broker name that bypasses name validation to include an xbean binding that can be later used by a VM transport to load a remote Spring XML application. The attacker can then use the DestinationView mbean to send a message to trigger a VM transport creation that will reference this malicious broker name which can lead to loading the malicious Spring XML context file. Because Spring's ResourceXmlApplicationContext instantiates all singleton beans before the BrokerService validates the configuration, arbitrary code execution occurs on the broker's JVM through bean factory methods such as Runtime.exec(). This issue affects Apache ActiveMQ: before 5.19.6, from 6.0.0 before 6.2.5; Apache ActiveMQ Broker: before 5.19.6, from 6.0.0 before 6.2.5; Apache ActiveMQ All: before 5.19.6, from 6.0.0 before 6.2.5. Users are recommended to upgrade to version 6.2.5 or 5.19.6, which fixes the issue.

The fix for CVE-2026-41635 was not applied to the 2.1.X and 2.2.X branches. Here was the original issue description: Apache MINA's AbstractIoBuffer.resolveClass() contains two branches, one of them (for static classes or primitive types) does not check the class at all, bypassing the classname allowlist and allowing arbitrary code to be executed. The fix checks if the class is present in the accepted class filter before calling Class.forName(). Affected versions are Apache MINA 2.1.0 <= 2.1.11, and 2.2.0 <= 2.2.6. The problem is resolved in Apache MINA 2.1.12, and 2.2.7 by applying the classname allowlist earlier. Affected are applications using Apache MINA that call IoBuffer.getObject(). Applications using Apache MINA are advised to upgrade.