Same-origin policy bypass in Google Chrome prior to 150.0.7871.125 enables remote attackers to manipulate cross-origin content integrity through the browser's HTML-in-Canvas rendering subsystem. Exploitation requires a victim to visit a crafted HTML page, making this a socially-engineered attack rather than a fully autonomous one. No active exploitation has been recorded (CISA KEV: absent, SSVC exploitation: none), and EPSS sits at 0.26% - indicating low real-world exploitation activity at time of analysis.
Deserialization logic in FasterXML jackson-databind 2.15.0 through 2.18.7, 2.21.3, and 3.1.3 allows unauthenticated remote callers to bypass @JsonIgnore annotations on Java Record types when a PropertyNamingStrategy is configured. The renamed JSON key - e.g., 'internal_role' produced by SnakeCaseStrategy from 'internalRole' - is not registered in the ignore list, so Jackson's constructor-parameter binding accepts the attacker-supplied value despite the developer's explicit exclusion directive. No public exploit tooling has been identified at time of analysis, but the upstream fix PR includes a runnable proof-of-concept test that fully demonstrates the bypass, lowering the bar for independent reproduction.
Windows Cryptographic Services improperly validates certificates across multiple Windows 10, 11, and Server editions, enabling remote unauthenticated attackers to bypass security features enforced by the PKI/certificate trust chain over a network path. Affected platforms span Windows 10 21H2 and 22H2, Windows 11 24H2 through 26H1, and Windows Server 2022 and 2025 in both full and Server Core deployments. No public exploit identified at time of analysis; EPSS at 0.24% (15th percentile) and SSVC exploitation status of 'none' together indicate low near-term exploitation probability despite the broad product footprint.
Denial of service in Roundcube Webmail's TNEF decoder allows remote senders to crash the webmail parser by embedding a crafted compressed-RTF size field inside a winmail.dat attachment. Affected versions are all Roundcube Webmail releases before 1.6.17 and 1.7.x before 1.7.2. Release notes confirm the defect manifests as an infinite loop in the TNEF parser, meaning repeated delivery of malicious emails could degrade server availability for affected users. No public exploit has been identified at time of analysis, and this vulnerability is not listed in CISA KEV.
Eclipse KUKSA Databroker 0.6.1 panics in a Tokio worker thread when an authenticated client sends a PublishValueRequest with a valid signal_id but omits the optional data_point field, causing the server to call Rust's unwrap() on a None value. Any client holding a valid JWT token can trigger this condition, cancelling the individual gRPC call while leaving the Databroker process intact and available. No public exploit has been identified and this vulnerability is not listed in CISA KEV; EPSS data was not supplied in available intelligence.
Incorrect access control in CAXPerts UniversalPlantViewer WebServices Server v2.7.6 allows any authenticated low-privilege user to invoke the /api/License/deactivateOffline endpoint and strip the server's license, rendering the service inoperable. The flaw (CWE-284) grants low-privileged accounts an action that should be restricted to administrative roles, providing a trivial denial-of-service primitive against industrial plant visualization infrastructure. No public exploit identified at time of analysis, though a researcher disclosure post on Medium appears to document the discovery.
Local tampering via symbolic-link following in Microsoft .NET (versions 8.0, 9.0, and 10.0) and the bundled toolchain in Visual Studio 2022 (17.12, 17.14) and Visual Studio 2026 (18.7) allows an authorized local attacker to redirect a privileged file operation to an unintended target, corrupting or replacing files outside their normal permissions. Microsoft (the reporter) has released a fix; there is no public exploit identified at time of analysis and the issue is not listed in CISA KEV. The high CVSS 3.1 base score of 7.0 is tempered by high attack complexity and the requirement for existing low-level local access.
OS command injection in Milestone Systems XProtect Management Server API enables authenticated users with edit permissions to execute arbitrary code as the Management Server Service. The CWE-78 flaw is network-accessible (AV:N) and requires no user interaction beyond possessing a privileged account, with CVSS 4.0 scope change metrics indicating high downstream impact (SC:H/SI:H/SA:H) to the broader surveillance infrastructure. No public exploit code or CISA KEV listing has been identified at time of analysis; Milestone has released patched versions addressed in their official advisory.
Cross-site scripting (CWE-79) in Microsoft Power BI Report Server lets an authenticated, low-privileged attacker inject script that executes in another user's browser session, enabling spoofing of report content and hijacking of the victim's authenticated context over the network. Exploitation requires a victim to interact with attacker-controlled report content (UI:R). There is no public exploit identified at time of analysis and the issue is not listed in CISA KEV; a vendor patch is available per the MSRC advisory.
Token exposure in the Auth0 Symfony SDK (auth0/symfony) versions 5.0.0-BETA0 through 5.8.0 allows OAuth 2.0 bearer access tokens to be submitted via a URL query parameter (`?token=`) in addition to the standard Authorization header, violating RFC 6750 Section 2.3 guidance. Any Symfony application using the Authorizer security authenticator is affected, with tokens silently recorded in server access logs, CDN and proxy logs, browser history, and Referer headers - all outside the intended OAuth token trust boundary. No public exploit has been identified and the vulnerability is not in CISA KEV; however, the design flaw is trivially reproducible by any authenticated user and creates persistent, passive token leakage risk.
Arbitrary file read in Apache OpenMeetings 5.0.0 through 9.0.x allows any room moderator to exfiltrate files accessible to the OS account running the server, including credentials and secrets, via a crafted download request to the WhiteBoard download service. The vulnerability stems from insufficient path validation in the WB download endpoint, enabling directory traversal outside the intended file scope. No active exploitation has been confirmed (not in CISA KEV), but the ability to read credential files makes this high-priority for any internet-exposed deployment where moderator access is not tightly controlled.
Stored XSS in osTicket's Agent and Admin panel allows any authenticated user with ticket-submission privileges to inject persistent JavaScript that executes in privileged staff browser sessions. Affected versions span two release branches: 1.10 through 1.17.7 and 1.18.0 through 1.18.3. No public exploit identified at time of analysis per KEV, but publicly available exploit code exists on GitHub (WesWrench/CVE-2026-36214), and the CVSS scope-change rating (S:C) reflects that successful exploitation compromises Agent or Admin sessions rather than the attacker's own context.
Privilege escalation via use-after-free in the Windows USB Print Driver affects Windows 11 (versions 24H2, 25H2, 26H1) and Windows Server 2025, requiring physical access to exploit. An attacker with hands-on access to a target machine can trigger a memory corruption condition through the USB print subsystem to achieve full local privilege escalation - High confidentiality, integrity, and availability impact per CVSS. No public exploit code has been identified at time of analysis, and this vulnerability is not listed in the CISA KEV catalog.
Unsafe deserialization in NVIDIA TensorRT-LLM's visual gen server through version 1.3.0 rc11 allows a locally privileged attacker to submit malicious payloads over the zeroMQ channel and achieve arbitrary code execution. The vulnerability stems from the visual gen server accepting and deserializing zeroMQ messages without enforcing adequate authorization controls, exposing CWE-502 deserialization risks. No public exploit has been identified at time of analysis, EPSS rates exploitation probability at 0.22% (12th percentile), and SSVC confirms no known active exploitation - though technical impact is rated total given full C/I/A compromise potential.
Stored Cross-Site Scripting in WP Customer Area WordPress plugin (versions up to and including 8.3.5) allows authenticated attackers with Contributor-level access or above to persistently inject arbitrary JavaScript via the unsanitized 'type' attribute of the customer-area-protected-content shortcode. The injected payload executes in the browser of any user who visits the compromised page, enabling session hijacking, credential theft, or malicious redirects against site visitors. No public exploit code has been identified and this CVE is not listed in the CISA KEV catalog at time of analysis; however, the Contributor-level prerequisite is a low barrier in multi-author WordPress environments.
Missing authentication for a critical function in NVIDIA TensorRT-LLM for Linux (all versions through v1.3.0 rc12) allows a local attacker with high privileges to bypass internal authentication controls and trigger unauthorized execution of critical operations. Successful exploitation leads to code execution, data tampering, and information disclosure against the LLM inference runtime environment. No public exploit code has been identified and SSVC assessment confirms no known active exploitation, though the technical impact is rated total by CISA's decision framework.
Time-of-check time-of-use (TOCTOU) race condition in the TraceEvent() system call of BlackBerry QNX Neutrino RTOS permits local attackers holding the PROCMGR_AID_TRACE kernel capability to trigger information disclosure, data tampering, or a kernel crash across three distinct product lines including safety- and medical-certified variants. Affected deployments span QNX Software Development Platform 7.0-7.1, QNX OS for Safety through 2.2.8, and QNX OS for Medical through 2.0.2. No public exploit or active exploitation has been identified; EPSS places this in the 1st percentile for near-term in-the-wild exploitation probability, and SSVC confirms exploitation status as none, though SSVC technical impact is rated total reflecting the kernel-level blast radius when conditions are met.
Stored Cross-Site Scripting in the News Kit Addons For Elementor WordPress plugin (versions ≤1.4.6) permits authenticated contributors to persistently inject arbitrary JavaScript into pages via the Site Logo Title and Single Author Box widgets by manipulating the elementor_ajax AJAX save request to bypass client-side tag-name restrictions. Every user who subsequently visits an injected page executes the attacker's script in their browser, enabling session hijacking, credential theft, or admin-level actions on their behalf. No public exploit has been identified at time of analysis and the vulnerability is not listed in the CISA KEV catalog; a vendor-released patch is available in version 1.4.7.
Memory exhaustion in elixir-mint's HTTP/2 client library (mint) allows a malicious or attacker-controlled HTTP/2 server to crash any BEAM application using the library by streaming unbounded zero-length CONTINUATION frames that bypass the existing byte-size guard. Affected versions span mint 0.1.0 through 1.9.1; a vendor-released patch is confirmed in 1.9.2. No public exploit or CISA KEV listing exists at time of analysis, but the explicit SSRF and redirect attack vectors meaningfully increase realistic exposure for server-side Elixir applications that proxy user-supplied URLs.
HTTP request smuggling in Eclipse Grizzly before 5.0.2 stems from the framework's inability to correctly parse malformed trailer header lines in chunked HTTP requests, enabling CWE-444 boundary-confusion attacks. Remote unauthenticated attackers who can send crafted chunked requests through a front-end proxy to a GlassFish-backed server can cause the proxy and Grizzly to disagree on request boundaries, smuggling attacker-controlled content as the prefix of a subsequent legitimate user's request. No public exploit code and no CISA KEV listing have been identified at time of analysis, though the CVSS 4.0 AT:P condition signals that specific deployment prerequisites must be met.
Server-side request forgery in Simple Machines Forum's image proxy allows any authenticated forum member to force the SMF host to make arbitrary internal HTTP requests by embedding attacker-controlled URLs in BBCode image tags. Because SMF automatically generates HMAC signatures authorizing any image URL the proxy encounters, an attacker can obtain a valid signed request targeting cloud instance metadata endpoints (169.254.169.254), RFC1918 internal web applications, and container network services - with no public exploit code identified at time of analysis and no CISA KEV listing. The highest-impact exposure is in cloud-hosted deployments where IMDS credential theft via SSRF could enable lateral movement beyond the SMF host itself.
Unauthorized command execution on NETGEAR DGND3700v1 routers (firmware ≤ V1.0.0.17) is achievable by any unauthenticated attacker present on the same local WiFi or LAN segment, exploiting improper input validation that effectively bypasses authentication controls. The CVSS 4.0 vector (AV:A/PR:N/VC:H/VI:H/VA:H) reflects full confidentiality, integrity, and availability impact on the vulnerable device. No public exploit code or confirmed active exploitation exists; NETGEAR identified this issue and published a July 2026 security advisory, though testing was conducted only against a simulated firmware environment and has not been verified on physical production hardware.
MIME type restriction bypass in TYPO3 CMS 14.2.0-14.3.5 allows unauthenticated attackers to upload files of arbitrary MIME types through forms configured with FileUpload or ImageUpload elements, circumventing the allowedMimeTypes restriction entirely. The flaw originates from a lifecycle ordering defect in the server-side form framework: MimeTypeValidator is registered before concrete form definition properties are applied, so it is absent from the validation pipeline at runtime. No public exploit code and no CISA KEV listing have been identified at time of analysis, but the network-accessible, zero-prerequisite-auth nature of the bypass elevates practical risk wherever affected forms are publicly exposed.
Privilege escalation in the Windows USB Print Driver affects Windows 11 (versions 24H2, 25H2, 26H1) and Windows Server 2025, exploitable by a physically-present attacker with low-privilege credentials via a race condition in the driver's shared resource handling. The flaw requires both physical device access and precise timing to win the synchronization window, making exploitation non-trivial. No public exploit identified at time of analysis; EPSS stands at 0.17% (7th percentile), and SSVC rates exploitation as none - consistent with high attack complexity and the physical access prerequisite.
Improper control of code generation in NVIDIA TensorRT-LLM for Linux (all versions through v1.3.0 rc12) allows a locally authenticated, high-privileged attacker to influence the code generation pipeline under complex, interaction-dependent conditions, potentially leading to arbitrary code execution, data tampering, and information disclosure. Despite the high CIA impact ratings, the attack prerequisites - local access, high privileges, high complexity, and required user interaction - severely constrain real-world exploitability. No public exploit exists, EPSS sits at the 2nd percentile (0.12%), and CISA SSVC records exploitation status as none.
Local information disclosure in Windows DirectX exposes sensitive data to unauthorized local actors across a broad range of Windows 10, Windows 11, and Windows Server versions. The vulnerability (CWE-200) achieves high confidentiality impact per the CVSS vector while producing no integrity or availability impact, meaning an attacker gains read access to protected data without disrupting system operation. No public exploit has been identified at time of analysis; EPSS at 0.38% (30th percentile) and SSVC exploitation status of 'none' collectively indicate low near-term exploitation probability despite the wide affected surface.
Windows Kernel information disclosure across a broad range of Microsoft Windows desktop and server editions allows a local, unprivileged attacker to read sensitive kernel memory or system state not intended for user-space access. The CVSS:3.1 vector AV:L/AC:L/PR:N/UI:N confirms local code execution with no special privileges is sufficient to trigger the flaw. EPSS sits at 0.38% (30th percentile) with no CISA KEV listing and SSVC exploitation status of 'none,' indicating no confirmed or publicly observed active exploitation at time of analysis; however, kernel information leaks of this class are routinely used to stage ASLR-bypass steps in multi-stage privilege escalation chains.
Open redirect in Adobe Commerce, Adobe Commerce B2B, Magento Open Source, and the Adobe Commerce Webhooks Plugin allows remote unauthenticated attackers to construct malicious URLs that silently forward victims to attacker-controlled phishing sites, enabling credential theft and account takeover. The flaw spans the full product family across multiple supported version branches and was disclosed by Adobe via Security Bulletin APSB26-73 with patches released in July 2026. No public exploit code has been identified at time of analysis, and SSVC rates exploitation status as 'none' with EPSS at 0.74% (50th percentile), indicating limited real-world exploitation signal despite broad product prevalence.
Improper input validation in Adobe's Content Credentials suite (C2PA Rust SDK, JS SDK, and c2patool CLI) allows a local attacker without privileges to crash the application by supplying malformed C2PA manifest content, resulting in a denial-of-service condition. Affected components span the Rust library (c2pa ≤ 0.84.0), the Node.js web SDK (@contentauth/c2pa-web ≤ 0.7.0), and the command-line tool (c2patool ≤ v0.16.5). No public exploit has been identified at time of analysis, and CISA SSVC rates exploitation status as none with non-automatable attack paths.
Integer underflow in Adobe's C2PA Content Credentials SDKs (Rust, JavaScript/Node.js, and CLI tool) allows an attacker to crash the host application by supplying a maliciously crafted C2PA manifest, resulting in a denial-of-service condition. All three release tracks are affected - the Rust SDK (≤0.84.0), the JS SDK (≤0.7.0), and c2patool (≤0.16.5) - with vendor-released patches available for each. No public exploit code has been identified at time of analysis, and EPSS sits at the 5th percentile, consistent with SSVC's 'exploitation: none' assessment.
Uncontrolled resource consumption in Adobe's C2PA Content Credentials Rust SDK, CLI tool (c2patool), and JavaScript web SDK allows a local attacker supplying a crafted input to exhaust system resources and crash the application. All three library families are affected across distinct runtimes - Rust, Node.js, and native binary - with patched releases available for each. No public exploit exists, SSVC classifies exploitation as none, and EPSS sits at the 5th percentile, indicating low real-world exploitation pressure despite the zero-user-interaction requirement.
Integer overflow in Adobe's Content Credentials SDK family (c2pa Rust SDK, c2pa-web JS SDK, and c2patool CLI) allows a local attacker to crash any application that processes a specially crafted C2PA manifest, resulting in denial of service. All three components share the vulnerable parsing logic and are affected across versions up to c2pa-v0.84.0, @contentauth/c2pa-web@0.7.0, and c2patool-v0.16.5 respectively. No public exploit exists and SSVC confirms no observed exploitation; however, patches are available and should be applied given the zero-privilege-required local trigger condition.
Integer underflow (CWE-191) in Adobe's Content Credentials ecosystem - spanning the Rust SDK (c2pa), JavaScript/Node.js SDK (c2pa-web), and c2patool command-line utility - crashes the application when processing a maliciously crafted C2PA manifest or media file, resulting in a denial-of-service condition. All three components share the same vulnerable code path, broadening the attack surface across toolchains. No public exploit has been identified at time of analysis; EPSS at 0.15% (5th percentile) and SSVC exploitation status of 'none' confirm this is not currently being targeted.
Unbounded GPU resource allocation in NVIDIA TensorRT-LLM's OpenAI-compatible inference API allows an attacker to exhaust GPU resources on the host system, resulting in denial of service. Affected versions span all releases up to and including v1.3.0 rc14. No public exploit or active exploitation has been identified at time of analysis; SSVC classifies exploitation as none and technical impact as partial, consistent with a localized availability-only impact.
Denial of service in NVIDIA TensorRT-LLM (versions through v1.3.0 rc15) allows an attacker who can submit requests to the OpenAI-compatible inference API to trigger a reachable assertion in the sampler thread, crashing the inference service. The impact is limited strictly to availability - no confidentiality or integrity compromise is possible. No public exploit has been identified, EPSS sits at 0.12% (2nd percentile), and CISA SSVC rates exploitation status as none, placing this at low operational priority for most teams.
NVIDIA TensorRT-LLM versions through v1.3.0 rc14 exposes a gRPC server chat API endpoint susceptible to improper input validation (CWE-20), enabling a local attacker to crash the inference service and cause denial of service. The impact is strictly limited to availability - no code execution, confidentiality breach, or integrity compromise is possible. No public exploit code exists, EPSS sits at the 2nd percentile (0.12%), and CISA SSVC rates exploitation as 'none' with technical impact as 'partial', collectively indicating low urgency despite the moderate CVSS 6.2 score.
Kernel crash in BlackBerry QNX Neutrino is triggerable by any local, unprivileged user via a malformed SchedGet() system call, resulting in full system availability loss. Affected platforms include QNX Software Development Platform 7.0 and 7.1, QNX OS for Safety (multiple branches up to 2.2.8), and QNX OS for Medical 2.0.2 and earlier. No active exploitation has been identified (EPSS 0.11%, CISA SSVC exploitation: none), but the safety-critical operational contexts in which QNX is deployed - automotive, medical devices, industrial automation - make a kernel crash disproportionately consequential relative to the moderate CVSS score.
Open redirect in Ivanti Xtraction before 2026.2.1 enables remote unauthenticated attackers to craft URLs that silently forward users to arbitrary external destinations. The vulnerability carries a Scope:Changed rating (S:C) in the CVSS vector, reflecting that the impact extends beyond the vulnerable application to the user's browser environment - a hallmark of effective phishing and credential harvesting campaigns. No public exploit code and no CISA KEV listing exist at time of analysis, and the AC:H metric in the vendor-provided vector suggests exploitation may not be trivially reliable in all configurations.
Reflected cross-site scripting in SourceCodester Class and Exam Timetabling System 1.0 allows remote unauthenticated attackers to inject arbitrary JavaScript into the browsers of users who interact with a crafted URL targeting the `day` parameter of /exam.php. A publicly available proof-of-concept exploit exists on GitHub (AlbaDove/cve issue #7), confirming exploitability, though no active exploitation has been confirmed in CISA KEV. The CVSS 4.0 base score of 2.1 and EPSS of 0.36% (28th percentile) together indicate low real-world exploitation probability, consistent with SSVC's finding that the attack is not automatable and technical impact is only partial.
Out-of-bounds read in Microsoft Excel's spreadsheet parser (CWE-125) exposes affected systems to both local information disclosure and application crash when a victim opens a specially crafted file, spanning Excel 2016 through Office LTSC 2024 and Office 365 for Mac. The CVSS vector (AV:L/UI:R/A:H) reveals that while the CVE description emphasizes information disclosure, the primary real-world impact is application availability - the invalid memory read can crash Excel (A:H), with incidental memory content leakage (C:L) as a secondary effect. No public exploit code has been identified at time of analysis and CISA SSVC assigns exploitation status of none, consistent with a low EPSS score of 0.31% at the 23rd percentile.
Path traversal in GoClaw 3.13.3-beta.3 allows authenticated remote attackers to read or write files outside the intended workspace directory by manipulating the path argument to the `writeFile` function in the ACP ToolBridge Workspace Handler (`internal/providers/acp/tool_bridge.go`). A proof-of-concept exploit has been publicly disclosed via a GitHub issue, lowering the bar for exploitation. No CISA KEV listing indicates active widespread exploitation has not been confirmed at time of analysis.
Reflected or stored cross-site scripting in Fortinet FortiOS, FortiPAM, and FortiProxy web interfaces allows a remote unauthenticated attacker to execute JavaScript in the browser session of an authenticated administrator via crafted HTTP requests. The CVSS temporal metric E:P confirms publicly available proof-of-concept exploit code exists, and the RL:O metric confirms Fortinet has released an official fix. No active exploitation has been confirmed by CISA KEV at time of analysis, but the combination of a POC and the high-value targets (network security appliances and privileged access management systems) elevates real-world risk beyond what the base CVSS score of 6.1 suggests.
Server-side request forgery in louisho5 picobot up to version 0.2.0 allows authenticated remote attackers to manipulate the url argument of the WebTool.Execute function in internal/agent/tools/web.go, causing the server to issue arbitrary HTTP requests to internal or external targets. The vulnerability carries a CVSS 4.0 score of 5.3 with low-privilege authentication required and has publicly available exploit code disclosed via a GitHub issue report. The upstream project has not responded to the responsible disclosure, meaning no patch is available at time of analysis.
Security feature bypass in the Windows System component on Windows 11 and Windows Server allows an authenticated local attacker with standard user privileges to circumvent an access control protection, yielding high confidentiality exposure and limited integrity impact. Affected platforms span Windows 11 versions 24H2 through 26H1 and Windows Server 2022 and 2025. No public exploit has been identified and CISA SSVC rates exploitation as none with non-automatable delivery, making this a moderate-priority local threat despite the high confidentiality impact signal in the CVSS vector.
Server-side request forgery in GoClaw 3.13.3-beta.3 enables authenticated remote attackers to manipulate the `output.video_url` parameter within the `bytePlusDownloadVideo` function, coercing the server into issuing HTTP requests to attacker-controlled destinations including internal network resources. The flaw resides in `internal/tools/create_video_byteplus.go` at the invoke endpoint. A proof-of-concept exploit has been publicly disclosed via GitHub issue #1199; however, no CISA KEV listing confirms mass active exploitation, and the CVSS 4.0 score of 2.1 reflects the low-privilege authentication barrier and limited per-request impact scope.
Missing authentication enforcement in the Microsoft Windows DNS component allows a low-privileged local attacker to tamper with DNS configuration or records without requiring elevated credentials. Affected platforms span Windows 10 (versions 1809 through 22H2), Windows 11 (versions 24H2 through 26H1), and Windows Server 2019 through 2025 - a broad deployment footprint across enterprise environments. No public exploit code has been identified and CISA SSVC rates exploitation as 'none' with partial technical impact, but the high integrity impact (CVSS I:H) warrants attention in multi-user environments and on DNS server infrastructure where record integrity is critical.
SQL injection in itsourcecode Electronic Judging System 1.0 exposes the admin panel endpoint /intrams/admin/add_judges.php to database manipulation via the unsanitized `fname` parameter. Authenticated remote attackers can craft malicious input to read, alter, or potentially enumerate backend database contents. A publicly available proof-of-concept exploit exists on GitHub, though the vulnerability is not listed in CISA KEV and carries a CVSS 4.0 base score of 2.1 (Low), reflecting the authenticated access prerequisite and limited-scope CIA impact.
Prototype pollution in compromise (spencermountain/compromise) through version 14.15.1 exposes all JavaScript applications using the library to Object.prototype contamination via the nlp.extend() Public Root API. A low-privileged remote attacker who can supply a crafted plugin argument containing __proto__, constructor, or prototype keys can inject properties into the global Object.prototype, producing partial confidentiality, integrity, and availability impact across the Node.js runtime. A public exploit exists via GitHub issue #1208, a vendor patch has been released (commit b4644ab7), and no CISA KEV listing has been confirmed at time of analysis.
Link-following vulnerability in louisho5 picobot up to version 0.2.0 allows a remote, low-privileged attacker to traverse outside the intended workspace directory via the CreateSkill and GetSkill functions in the filesystem handler. The root cause is improper symlink resolution (CWE-59) in internal/agent/tools/filesystem.go, enabling potential unauthorized file read and write operations on the host filesystem. A public exploit exists as a GitHub issue report; no vendor patch has been released and the project maintainer has not yet responded to the disclosure.
Server-side request forgery in zhayujie CowAgent's Vision Tool allows authenticated remote attackers to make the server issue arbitrary HTTP requests to internal network resources. The vulnerability exists in the Vision._download_to_data_url / _build_image_content function in agent/tools/vision/vision.py, where the image argument is passed to an HTTP fetch without URL validation, enabling access to private RFC1918 addresses, loopback interfaces, and other non-public infrastructure. A publicly available proof-of-concept exploit exists (GitHub issue #2878), and a vendor-released patch (version 2.1.2) is available.