Path traversal in gitoxide before 0.52.1 and gix before 0.83.0 allows an attacker who supplies a malicious git repository to redirect submodule git directory resolution outside .git/modules, exposing arbitrary on-disk repository structure to inspection. The flaw stems from submodule names read from .gitmodules being joined directly as filesystem path components with no normalization or confinement check, so a name such as '../../../escaped-target.git' escapes the expected directory boundary. A proof-of-concept is publicly available via GHSA-fr8x-3vfx-f45h; no confirmed active exploitation at time of analysis.
Path traversal in Qwen-Agent through 0.0.34 allows unauthenticated remote attackers to read arbitrary files from the server filesystem by supplying absolute paths to the Gradio document-parsing interface. The flaw exists in the simple_doc_parser and supporting utility code, which pass caller-controlled paths directly to file-open operations without restricting access to an intended directory. The CVSS 4.0 score of 8.7 (AV:N/AC:L/PR:N) reflects a zero-prerequisite, network-accessible attack surface; no active exploitation is confirmed in CISA KEV and no public POC has been identified at time of analysis.
Denial of service in SvelteKit versions 2.49.0 through 2.53.2 is achievable by unauthenticated remote attackers who submit crafted multipart form requests that exploit a deserialization expansion flaw in the experimental `form` remote function, causing the server to construct disproportionately large `files` arrays and exhaust processing resources. Exploitation is gated behind a non-default opt-in flag (`experimental.remoteFunctions: true`) and requires the application to process the `files` array without length or size validation, substantially narrowing real-world exposure despite the CVSS 4.0 score of 8.7. No public exploit has been identified at time of analysis and the vulnerability does not appear in the CISA KEV catalog.
Denial-of-service in gitoxide's gix-pack crate before 0.69.0 exposes Rust Git clients to remote process crashes during clone or fetch operations via two distinct attack vectors. Unchecked array indexing in the delta application routine causes index-out-of-bounds panics on truncated pack data, while uncapped buffer allocation from attacker-controlled LEB128 size headers allows a 10-byte crafted pack entry to trigger an immediate out-of-memory process kill - both exploitable by any Git server the victim connects to. No active exploitation has been confirmed (not in CISA KEV), but the unauthenticated, network-reachable attack surface with trivial complexity makes this a high-priority patch for any environment running automated clone or fetch workflows against external remotes.
CPU exhaustion in SvelteKit's experimental remote form deserialization endpoint allows unauthenticated remote attackers to render affected servers unresponsive by submitting malformed form data, resulting in denial of service. Versions 2.49.0 through 2.52.1 of @sveltejs/kit are affected, but only when both the experimental.remoteFunctions and form features are simultaneously enabled - a non-default, opt-in configuration. No public exploit code or active exploitation (CISA KEV) has been identified at time of analysis; the CVSS 4.0 score of 8.7 reflects the complete availability impact against network-reachable endpoints with no authentication barrier once the experimental features are active.
Memory exhaustion in @sveltejs/kit versions 2.49.0 through 2.52.1 allows unauthenticated remote attackers to crash the server process by sending malformed form data to a remote function endpoint, resulting in complete denial of service. Exploitation is gated behind two simultaneous non-default opt-in conditions - the experimental.remoteFunctions flag and form usage - which significantly narrows the real-world attack surface relative to the CVSS 4.0 score of 8.7. No public exploit or CISA KEV listing has been identified; a vendor-released patch is available in version 2.52.2.
Stack-based buffer overflow in the MongoDB BI Connector ODBC driver allows a low-privileged SQL-capable user to overwrite adjacent stack memory by submitting a positioned-cursor statement with a cursor name exceeding the size of an internal fixed-length buffer. Any application exposing MongoDB through this ODBC driver - including common BI tools such as Tableau, Excel, or MicroStrategy - is affected across all currently listed versions (CPE wildcard). Successful exploitation can crash the hosting application process and may enable unintended code execution within that process context; no public exploit or CISA KEV listing is confirmed at time of analysis.
Server-side request forgery in Portkey AI Gateway through version 1.15.2 enables unauthenticated remote attackers to weaponize the gateway as an internal network proxy by injecting arbitrary host values into the x-portkey-custom-host header on the /v1/proxy/* route, which lacks requestValidator middleware. The gateway faithfully forwards outbound requests - including Authorization headers carrying LLM provider API keys - to attacker-specified internal addresses, enabling credential exfiltration and internal service enumeration. No public exploit code has been identified at time of analysis and no CISA KEV listing exists, but the unauthenticated, low-complexity nature of the attack vector makes this an urgent remediation priority for any internet-exposed gateway deployment.
Authenticated OS command injection in PLANET GS-4210-16P2S managed switch firmware prior to 3.441b260626 enables a remote administrator to execute arbitrary commands on the device operating system via a crafted certificate upload request targeting /cgi-bin/httpuploadcert.cgi. The certificate password field is incorporated into a shell command without sanitization of shell metacharacters, a textbook CWE-78 root cause. No public exploit code or CISA KEV listing has been identified at time of analysis; exploitation requires valid administrator-level web credentials, confirmed by CVSS 4.0 PR:H.
Log storage exhaustion in MongoDB Connector for BI (mongosqld) allows any unauthenticated network peer to crash the shared process by generating sufficient connection log volume to fill the configured log path's backing storage. When a subsequent log write or rotation operation fails, an uncaught exception (CWE-248) terminates the mongosqld process, immediately denying SQL service to all connected clients. The process cannot recover autonomously - it continues to fail on startup until an operator manually reclaims disk space - and the diagnostic message explaining the condition is never persisted, compounding operational triage difficulty. No active exploitation has been identified at time of analysis.
Command injection in Snyk Sweater Comb before version 3.8.8 allows an attacker who controls the .vervet.yaml configuration file to execute arbitrary OS commands on the host running the lint operation. The root cause is the expectGitBranch() function in src/lint.ts passing the unsanitized linters.<key>.optic-ci.original branch name value directly into Node.js child_process.exec() via an unescaped template literal, where the shell interpreter evaluates injected metacharacters as commands. The highest-risk exploitation path is a malicious pull request that plants a crafted .vervet.yaml, triggering command execution when a maintainer or CI pipeline subsequently runs sweater-comb lint; no CISA KEV listing or confirmed public exploit exists at time of analysis.
Stack-based buffer overflow in TP-Link TL-MR100 V3.20 allows an adjacent unauthenticated attacker to corrupt the httpd process stack by sending a crafted encrypted payload to the /cgi/login endpoint before any authentication occurs. The flaw resides in the http_gdpr_decrypt function, which performs insufficient bounds checking on incoming data, enabling overwrite of saved control-flow data and potential arbitrary code execution in the context of the httpd process. Vendor patch is available; no public exploit code or CISA KEV listing has been identified at time of analysis.
Unauthenticated connection exhaustion in MongoDB Connector for BI allows any network-reachable client to hold authentication sessions open indefinitely by initiating a SASL login exchange and then withholding further data. The absence of a wall-clock timeout on the negotiation loop and no read deadline on the client socket means each such session permanently occupies a worker thread, a connection slot, and associated backend MongoDB handles until the process is restarted. Sustained repetition of this pattern can exhaust the configured connection pool capacity entirely, denying service to all legitimate BI tool users; no public exploit has been identified at time of analysis.
Hardcoded HTTP Digest authentication credentials baked into CP Plus CP-XR-DE21-S Router firmware grant any adjacent-network attacker full administrative access without valid credentials. Because the credentials are identical across every device running the affected firmware, a single firmware extraction exposes the entire deployed population. No active exploitation is confirmed by CISA KEV, but the trivially low exploitation complexity and total administrative compromise make this a high-priority remediation target for any network where this router is deployed.
Mongosqld in the MongoDB BI Connector requests but does not enforce client certificates during the TLS handshake when configured with a CA file, silently accepting connections from clients that present no certificate at all. Deployments that rely on client certificate authentication as the sole means of identifying users are fully exposed - any remote party with network access to the listener can establish a session and read all MongoDB data surfaced through the SQL interface. The CVSS 4.0 score of 8.7 reflects the network-accessible, zero-privilege nature of the flaw; no public exploit code or CISA KEV listing has been identified at time of analysis, and real-world impact is bounded by a specific non-default configuration prerequisite.
Credential disclosure in AUTOMATIC1111 Stable Diffusion WebUI through version 1.10.1 exposes configured authentication credentials to unauthenticated network attackers via the `/sdapi/v1/cmd-flags` REST endpoint. The endpoint serializes all runtime command-line arguments - including `gradio_auth` and `api_auth` values - into a cleartext JSON response with no access control, directly nullifying the authentication protection those flags are meant to provide. No public exploit or active exploitation has been confirmed, but the attack path is a single unauthenticated HTTP request to a known endpoint, making this a high-priority finding for any publicly reachable deployment.
Remote code execution in BISHENG's workflow engine (versions before 2.6.0) allows any authenticated user to submit crafted Code node definitions to the POST /api/v1/workflow/run_once endpoint, where they are evaluated via Python's exec() built-in without sandboxing or process isolation. Successful exploitation grants full server-side code execution with access to the host filesystem, stored credentials, and internal network resources reachable from the BISHENG backend process. No public exploit or CISA KEV listing has been identified at time of analysis, but the low-complexity, authenticated network attack path represents a critical risk especially in multi-tenant or shared-access deployments.
Server-side request forgery in Qwen-Agent through 0.0.34 allows unauthenticated remote attackers to exfiltrate data from internal infrastructure by exploiting the document parsing path's failure to validate caller-supplied URLs. The Gradio interface - exposed without authentication by default - serves as the entry point, enabling attackers to direct the server to issue HTTP requests to cloud metadata endpoints such as AWS IMDSv1 or GCP metadata APIs and retrieve responses through the document parsing output. No public exploit code or CISA KEV listing exists at time of analysis, but the zero-privilege, zero-interaction attack vector against credential-bearing metadata services makes this a high-priority risk for any cloud-hosted deployment.
Credential exfiltration in gitoxide's gix-transport crate allows a server controlling an HTTP redirect response to steal the caller's HTTP Basic Authorization credentials by exploiting a hand-rolled URL parser in gix-url that misidentifies the authority component of crafted redirect URLs. Affected are gix-url versions ≤ 0.32.0 and gix-transport versions ≤ 0.49.0; the redirect identity guard `can_reuse_identity` fails open when the parser does not treat '?' or '#' as authority terminators per RFC 3986, causing credentials to be forwarded to an unintended host. No public exploit has been identified at time of analysis, though the CVSS 4.0 score of 8.7 reflects a full confidentiality impact with no authentication prerequisites.
Unauthenticated blind SQL injection in the Sexy Polling Reloaded Joomla extension (Jefferson49) affects all versions below 5.6.1, enabling remote attackers to exfiltrate the entire underlying database without credentials. The CVSS 4.0 score of 8.7 reflects a trivial exploitation path - no authentication, no user interaction, and no configuration preconditions required against default installations of the extension. No public exploit code or CISA KEV listing has been identified at time of analysis, though the fully unauthenticated network vector makes opportunistic automated scanning a realistic threat vector.
CORS misconfiguration in Rybbit before 2.7.0 allows any attacker-controlled website to issue credentialed cross-origin requests to the Rybbit analytics server, because the server unconditionally reflects the request's Origin header in Access-Control-Allow-Origin while simultaneously enabling Access-Control-Allow-Credentials. Authenticated victims who visit a malicious page expose their analytics data, account information, and permit the attacker's script to perform state-changing API operations on their behalf. No public exploit has been identified at time of analysis, but the underlying attack class requires only standard browser Fetch API usage and is trivial to implement.
Unauthenticated OS command injection in openNDS before 11.0.0 allows any adjacent-network attacker - such as an unprivileged user connected to a captive portal WiFi - to execute arbitrary shell commands on the host router by supplying a crafted `fas` query parameter to the `/opennds_preauth/` endpoint. The vulnerability originates in `libopennds.sh`, which processes the parameter without sanitizing shell metacharacters before passing it to OS-level commands. The fix, published in commit 8c03750, introduces strict base64-character allowlist validation that rejects any query string containing characters outside A-Za-z0-9+/=. No public exploit code or CISA KEV listing has been identified at time of analysis.
Kernel panic vulnerability in the Linux kernel's IP fragment reassembly path allows an unprivileged user in their own user namespace to crash the host kernel by writing two specially crafted packets with GSO (Generic Segmentation Offload) metadata to a tun/tap device. The flaw exists because inet_frag_queue_insert() - shared by IPv4, IPv6, nf_conntrack_reasm, and 6lowpan reassembly - fails to strip GSO state from fragments before queuing, causing the reassembled skb to reach skb_segment() with SKB_GSO_DODGY state and trigger BUG_ON() assertions. The vulnerability is also reachable with CAP_NET_RAW in a netns where a defrag point precedes a GSO point, and from a guest VM whose VMM forwards virtio_net_hdr to a host tap; KEV status is unconfirmed and EPSS is low at 0.18% (7th percentile), though the CVE description contains a working reproducer.
Missing ownership validation across three Quivr chat API endpoints through version 0.0.322 enables any authenticated user to read, delete, or inject fabricated messages into other users' private conversations and associated knowledge base content. The vulnerable endpoints (GET /chat/{chat_id}/history, DELETE /chat/{chat_id}, POST /chat/{chat_id}/question/answer) accept user-supplied chat IDs without server-side authorization checks, a classic IDOR (CWE-639) pattern. No public exploit code has been identified and this CVE is not listed in the CISA KEV catalog at time of analysis, but the low-complexity, authenticated-only prerequisite makes exploitation trivially accessible to any malicious account holder in a multi-user deployment.
Insecure Direct Object Reference in VoltAgent through 2.1.20 allows any authenticated user to read, modify, or delete conversations belonging to other users by supplying arbitrary conversation identifiers to memory API endpoints. The server accepts caller-controlled identifiers without verifying ownership, enabling horizontal privilege escalation across all users of a shared deployment. No public exploit has been identified at time of analysis, but the IDOR pattern (CWE-639) is trivially exploitable by any account holder with basic API knowledge.
API authentication middleware in Gophish through 0.12.1 fails to enforce account lockout and mandatory password-change states, allowing any holder of a valid API key to retain full platform access regardless of administrative security actions taken on that account. Affected versions include all releases up to and including 0.12.1 per CPE cpe:2.3:a:gophish:gophish:*:*:*:*:*:*:*:*. An attacker who obtains or retains an API key - including a compromised insider or a key that was not rotated before an account was locked - can read and modify phishing campaign data, harvested credentials, and user configurations with high confidentiality and integrity impact. No public exploit or CISA KEV listing has been identified at time of analysis.
Privilege escalation in Budibase server before 3.41.3 allows an authenticated app-scoped builder to gain unauthorized builder access across all applications within the same tenant. By submitting crafted requests to the public user create or update API endpoints with manipulated builder.apps fields, the attacker bypasses the missing authorization check and self-assigns elevated builder privileges over unrelated applications. No public exploit code has been identified and no CISA KEV listing exists at time of analysis, but the network-accessible, low-complexity nature of the attack makes this a significant risk in any multi-tenant Budibase deployment.
Authorization bypass in Budibase Server before 3.41.3 allows any authenticated user holding the lowest-privilege BASIC role to perform unrestricted read, create, update, and delete operations against any application table via the POST /api/datasources/query endpoint. Per-table role restrictions enforced elsewhere in the application are entirely absent on this API path, rendering the access control model trivially circumventable with a standard API call. No public exploit has been identified at time of analysis, though the low attack complexity - requiring only valid BASIC credentials and a crafted HTTP request - makes this a genuine priority for externally-accessible or multi-tenant deployments.
SQL injection via identifier escaping failure in MongoDB Connector for BI allows a low-privileged MongoDB user to craft collection, field, or index names that break out of SQL quoted identifiers, embedding arbitrary SQL into the DDL output produced by SHOW CREATE statements. Any user with write access to a sampled MongoDB collection can plant the payload; the injected SQL executes only when an operator or automated pipeline later replays that generated DDL against a downstream SQL-compatible server, triggering execution with that session's privileges. No public exploit has been identified and this CVE is not listed in CISA KEV, though the attack requires only low privileges and exploits a plausible - and often automated - operational workflow in analytics environments.
SQL injection in WPBulky WordPress plugin (versions ≤ 1.2.2) is exploitable by any authenticated user holding a Contributor-level role, requiring no user interaction and no special configuration beyond plugin installation. The CVSS Scope:Changed flag combined with high confidentiality impact indicates the injection can reach beyond WordPress application data into the underlying MySQL database, enabling extraction of password hashes, private content, and any other data accessible to the database user. No public exploit code or CISA KEV listing has been identified at time of analysis, but the low-complexity, low-privilege attack vector makes this a realistic threat on sites with open user registration.
Out-of-bounds read in the Linux kernel's asus_atk0110 hardware monitoring driver allows a malformed ACPI GGRP firmware package to trigger kernel memory corruption. Systems running ASUS ATK0110-equipped hardware with a BIOS that returns zero-count sub-packages in its GGRP management group response are affected across a wide range of Long-Term Support and mainline kernel versions from 2.6.32 onward. Exploitation is contingent on a specific hardware/firmware defect condition rather than a direct user action, but the resulting out-of-bounds read in kernel space carries high potential impact across confidentiality, integrity, and availability. No public exploit or CISA KEV listing exists; EPSS is 0.18% (7th percentile), reflecting the niche hardware prerequisite.
Race condition and error-handling flaws in the i2c-imx I2C controller driver allow a NULL pointer dereference via the kernel interrupt handler on NXP i.MX SoC-based systems. The vulnerability affects multiple stable kernel branches (5.15 through 7.x) and is triggered when I2C slave registration fails partway through a PM runtime resume, leaving a stale or NULL pointer that the shared IRQ handler can dereference concurrently. The realistic impact is a kernel panic causing system unavailability; no public exploit exists and EPSS is 0.17%, consistent with no observed active exploitation.
The Linux kernel ims-pcu USB driver exposes sysfs device attributes on the secondary data interface in addition to the intended primary control interface, allowing local users to trigger kernel crashes or undefined behavior by accessing those attributes on the data interface, which lacks the descriptors and internal state the driver requires. Affected systems must have an IMS Passenger Control Unit USB device physically connected and run a kernel version between commit 204d18a7a0c67352857dee1bbac517ed63f01d8e and the respective fix commits. No public exploit code exists, and the EPSS score of 0.17% (6th percentile) reflects the extreme hardware rarity required for exploitation; no CISA KEV listing exists.
Race condition in the Linux kernel's hisi_sas v3 hardware driver triggers a kernel WARNING and potential memory corruption when SAS phy link reset and driver removal (rmmod) execute simultaneously. Specifically, during SAS phy bring-up, device_links_driver_bound sets the link status to DL_STATE_AVAILABLE, and a concurrent rmmod then causes __device_links_no_driver() to encounter an inconsistent device-link state. No public exploit exists and EPSS sits at 0.17% (6th percentile), reflecting this is a hardware-specific, privilege-intensive, timing-dependent flaw with very low exploitation likelihood in practice.
Out-of-bounds memory write in the Linux kernel's device tree reserved memory subsystem allows an attacker with control over device tree content to corrupt kernel memory at boot time, with theoretical full kernel compromise (C:H/I:H/A:H). The fdt_scan_reserved_mem() function writes past the end of a fixed-size local array when a device tree blob defines more dynamically-placed /reserved-memory subnodes than MAX_RESERVED_REGIONS allows. Despite a CVSS score of 8.4, exploitation is heavily constrained by the requirement to modify device tree content before boot - a capability requiring privileged firmware access - and EPSS at 0.16% (5th percentile) reflects negligible observed exploitation probability. No public exploit has been identified and the vulnerability is absent from CISA KEV.
The spi-qpic-snand NAND flash driver in Linux kernel 6.18 and later contains a command-ordering defect that causes every SET_FEATURE write to apply the previous operation's value rather than the intended one, producing an off-by-one effect. On Qualcomm IPQ5018-based hardware (confirmed on TP-Link Archer AX55 v1 with ESMT F50L1G41LB flash), this defect became destructive in v6.18 when SPI-NAND OTP support was introduced: the 'disable OTP mode' call erroneously leaves CFG_OTP_ENABLE set permanently, causing all subsequent flash reads to return OTP area content and all writes to fail with -EIO, rendering the device unbootable. No public exploit identified at time of analysis; EPSS of 0.15% (5th percentile) correctly reflects that this is a driver logic defect rather than a traditionally attacker-controllable vulnerability.
Restriction bypass in the Linux kernel's io_uring subsystem allows a local low-privileged user to shed per-task io_uring security restrictions by executing exec(). When a task that has established io_uring restrictions calls exec(), the kernel's exec cancellation path invokes __io_uring_free(), which incorrectly frees both the task context and the per-task restriction simultaneously. Any io_uring ring created by the post-exec process is then entirely unrestricted, defeating sandbox and policy enforcement intended to constrain io_uring operations. No public exploit has been identified at time of analysis, but the CVSS score of 8.4 with Changed Scope (S:C) reflects that the bypass can expose confidentiality and integrity of resources guarded by the dropped restrictions.
SiYuan before v3.8.1 exposes a DNS rebinding SSRF vulnerability in its http_request and web_fetch agent tools, allowing network-accessible attackers who can influence processed URLs to bypass the built-in SSRF IP-blocklist defense and reach cloud instance metadata services or internal network hosts. The guard-time-only DNS resolution creates a TOCTOU race: the SSRF check sees a legitimate public IP, but the actual TCP connection lands on 169.254.169.254 or another private address. No public exploit code or CISA KEV listing exists at time of analysis; however, the CVSS 4.0 score of 8.4 reflects high downstream confidentiality risk for cloud-hosted deployments where metadata credentials are accessible.
Command injection in oPanel's advanced/curl component allows authenticated attackers to execute arbitrary shell commands on the underlying server. The 'url' parameter accepted by the built-in curl tool is not sanitized, enabling shell metacharacter injection by any logged-in user. A public proof-of-concept is available at the linked GitHub repository, and SSVC rates the technical impact as total, meaning full system compromise is achievable post-authentication.
Server-side request forgery in Budibase Server before 3.41.3 allows authenticated builder-level users to supply arbitrary URLs to the datasource verify endpoint without SSRF validation, causing the server to make outbound requests to attacker-controlled infrastructure. The primary impact is credential leakage: internal CouchDB connection credentials are exposed in those outbound requests, granting an attacker direct, full administrative access to the backend database. This risk is amplified in cloud deployments where CouchDB serves as the primary data store. No public exploit code or CISA KEV listing is identified at time of analysis.
Schema-sampling denial-of-service in MongoDB Connector for BI allows a low-privileged database user to permanently halt the schema-refresh routine by creating a view in a sampled namespace that reliably fails on evaluation. The mongosqld process misclassifies the persistent failure as transient, exhausts its retry budget, then continues running with no valid schema - leaving all downstream SQL clients unable to retrieve results until an operator manually removes the offending view or excludes the namespace. No public exploit has been identified at time of analysis, but the low privilege requirement and network accessibility make this a credible insider or multi-tenant threat.
Server-Side Request Forgery in Gitingest through 0.3.1 allows unauthenticated attackers to redirect outbound server connections to arbitrary attacker-controlled hosts by exploiting a prefix-only hostname validation bypass in the `_validate_host` function. Because Gitingest may carry GitHub personal access tokens as HTTP basic auth credentials on outbound git requests, any host whose name begins with `git.`, `gitlab.`, or `github.` passes validation and becomes a credential harvesting endpoint. No public exploit code is confirmed at time of analysis and no CISA KEV listing exists, but the CVSS 4.0 score of 8.3 reflects the combination of low attack complexity, unauthenticated access, and high confidentiality impact.
Cross-application resource injection in Budibase Server before 3.41.3 allows any authenticated builder to plant tables, automations, queries, and screens into workspaces they have no authorized access to. The POST /api/resources/duplicate endpoint accepts an arbitrary destination workspace ID in the request body and performs no authorization check against the destination, enabling horizontal privilege escalation across workspace boundaries. Once malicious automations are injected, an attacker can trigger them to fire outbound webhooks relaying victim application data to attacker-controlled infrastructure, achieving data exfiltration. No public exploit has been identified at time of analysis, and a vendor-released patch is available in version 3.41.3.
Namespace collision between user identifiers in IBM Langflow OSS 1.0.0 through 1.11.1 exposes sensitive information and enables unauthorized message injection by unauthenticated remote attackers. The flaw, rooted in CWE-639 (authorization bypass via user-controlled key), allows an attacker to resolve or collide with another user's namespace without requiring authentication, as confirmed by the CVSS vector (PR:N). No public exploit code or CISA KEV listing has been identified at time of analysis, but the low-complexity, network-accessible attack surface warrants prompt remediation.
Remote unauthenticated execution of arbitrary AI workflows in IBM Langflow OSS 1.0.0 through 1.11.1 exposes sensitive information via an authentication bypass (CWE-287). The CVSS 3.1 base score of 8.2 (AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:L/A:N) confirms zero prerequisites - any network-reachable instance can be targeted without credentials. No public exploit code or CISA KEV listing has been identified at time of analysis, but the trivial attack complexity combined with high confidentiality impact and vendor-confirmed patch availability makes this an immediate remediation priority for any organization running an exposed Langflow instance.
Server-side request forgery in Logto through 1.42.0 allows tenant administrators holding Management API credentials to read responses from arbitrary internal network services by supplying attacker-controlled URLs as the OIDC SSO connector issuer parameter. The vulnerable surface extends beyond the SSO connector endpoint to webhook delivery and SAML IdP metadata fetching, each of which accepted tenant-supplied URLs without filtering loopback, RFC 1918, or cloud metadata (169.254.169.254) destinations. No public exploit identified at time of analysis; a patch is available upstream via commit 16f4b2e7.
Server-side request forgery in Logto through 1.42.0 allows tenant administrators holding Management API tokens to direct the server to issue HTTP POST requests to arbitrary internal URLs via the POST /api/hooks/:id/test webhook-test endpoint, then read response bodies from private-network services back through the API error payload. The flaw stems from the absence of host validation in the webhook delivery library (packages/core/src/libraries/hook/utils.ts) and extends beyond webhooks to enterprise SSO connector discovery and SAML metadata fetching. A patch is available via upstream commit 16f4b2e and no public exploit code or CISA KEV listing has been identified at time of analysis.
Kernel memory corruption path in Linux's tunnel metadata subsystem (`tun_dst_unclone`) triggers a FORTIFY_SOURCE memcpy overflow warning when the kernel is built with clang, due to the `__counted_by` annotation on a flexible array member being evaluated as zero at copy time. Systems using GENEVE tunneling on clang-built kernels with source fortification enabled may experience kernel availability disruption if the warning escalates to a panic. No public exploit has been identified; EPSS of 0.17% at the 7th percentile reflects low exploitation probability, and the vulnerability is not listed in CISA KEV.
Resource exhaustion in filebrowser through 2.63.23 allows denial-of-service by exploiting unvalidated named pipes in the archive and public download handlers. Both authenticated users and anonymous visitors holding public share links can trigger blocking open() syscalls by repeatedly requesting archives that include named pipe (FIFO) filesystem entries, pinning server goroutines until connection resources are exhausted. No public exploit has been identified at time of analysis, and no CISA KEV listing exists, but the anonymous attack path via public shares elevates the practical concern for internet-exposed deployments.
Cross-origin credential exposure in HeyForm before 3.0.0-rc.8 allows any attacker-controlled web page visited by a logged-in user to make fully authenticated GraphQL API requests against the HeyForm server. The server's default CORS configuration (`origin: true, credentials: true`) unconditionally echoed the request Origin header back as `Access-Control-Allow-Origin` while also asserting `Access-Control-Allow-Credentials: true`, defeating the browser's same-origin policy for all origins. This exposes workspaces, projects, forms, submissions, and respondent data to exfiltration and permits account settings modification; no public exploit has been identified at time of analysis.