Overzicht van binnengekomen advisories.
2179 resultaten gevonden
CVE-2026-65092 affects NVIDIA OpenShell Sandbox for Linux, where an attacker can exploit a path traversal vulnerability to bypass L7 REST network policy controls. The vulnerability could allow unauthorized access to sensitive information and enable data tampering. The flaw resides in the network policy enforcement layer, specifically the Layer 7 REST policy mechanism. Successful exploitation does not appear to require authentication but likely requires network access. NVIDIA has published a security advisory via their product-security GitHub repository. The vulnerability has been assigned a high criticality rating. Affected users are advised to review NVIDIA's advisory for patches or mitigations.
Bekijk origineel advisory →Alluxio's S3 REST proxy fails to verify AWS Signature Version 4 signatures in its default configuration, creating a critical authentication bypass vulnerability. Unauthenticated attackers can extract usernames from unsigned Authorization headers and impersonate any user, including privileged service accounts. This allows attackers to perform unauthorized read, write, and delete operations on arbitrary data. The vulnerability affects Alluxio through version 2.9.5. The flaw stems from the proxy not enforcing signature verification, meaning any attacker with network access can spoof user identities without valid credentials. This poses a significant risk to data integrity and confidentiality in environments using Alluxio's S3-compatible interface.
Bekijk origineel advisory →CVE-2026-79788 affects Dradis Community Edition versions 5.1.0 through 5.2.0. The ProvidersController and AgentsController incorrectly gate their admin_required authorization check on the constant Dradis::Pro, which is never defined in the CE edition, meaning the check is never enforced. Any authenticated non-admin user can exploit this to create an AI provider pointing to an arbitrary HTTP/HTTPS address, including internal or link-local hosts like http://169.254.169.254. When an AI interaction is triggered, the server makes a request to the attacker-controlled URL, constituting a Server-Side Request Forgery (SSRF). For non-2xx HTTP responses, the target's response body is reflected back to the attacker via ActionCable/Turbo Stream error messages, making the SSRF readable and exploitable for internal network reconnaissance. The vulnerability stems from a broken authorization logic flaw combined with insufficient URL validation.
Bekijk origineel advisory →CVE-2026-79784 describes a critical arbitrary code execution vulnerability in the Vocos Python library, a neural vocoder used in AI/ML workflows. The vulnerability exists in the instantiate_class function within vocos/pretrained.py, which loads a class path from a configuration file without any allowlist or restriction on which class may be named. The function uses Python's __import__ and getattr to dynamically import and instantiate any callable, passing user-controlled arguments. The Vocos.from_pretrained method downloads a remote config.yaml from a Hugging Face repository specified by the caller and passes it directly to from_hparams, meaning loading a model from an untrusted repository results in execution of arbitrary code chosen by the repository owner. This is distinct from the torch.load deserialization risk, which is mitigated in newer PyTorch versions. The attack surface is significant for any user or pipeline that loads Vocos models from third-party or community Hugging Face repositories. No allowlist or sandboxing constrains the class path resolution, making this a straightforward remote code execution vector during model loading.
Bekijk origineel advisory →A stack-based buffer overflow vulnerability has been identified in BlueZ, the Linux Bluetooth protocol stack. An attacker within Bluetooth radio range can exploit this flaw by sending a specially crafted Extended Inquiry Response (EIR) packet during Bluetooth device discovery. The vulnerability triggers a buffer overflow condition that can crash the bluetoothd service, resulting in a Denial of Service (DoS). Beyond DoS, the flaw may also allow for arbitrary code execution on the affected system. The attack requires no authentication and only proximity within Bluetooth range. This affects Linux systems running BlueZ during active Bluetooth discovery operations. The vulnerability has been acknowledged by Red Hat and tracked in their security advisory and Bugzilla systems. A GitHub security advisory has also been published by the BlueZ project maintainers.
Bekijk origineel advisory →A SQL injection vulnerability has been identified in SililaWijesinghe's Food Ordering System, affecting the /search.php file. The vulnerability is triggered by manipulating the 'search_box' argument, allowing remote attackers to perform SQL injection attacks. The affected version spans up to commit ba314e897e3365600461e5ea59432e39ceaa0fa5. The product uses a rolling release model, making specific version identification difficult. The exploit has been publicly disclosed and is available for use. Remote exploitation is possible, increasing the risk of unauthorized database access or data exfiltration. The vendor was notified prior to public disclosure but did not respond. No patch or mitigation has been officially confirmed. The public availability of the exploit elevates the urgency for users of this system to take protective measures.
Bekijk origineel advisory →A missing authentication vulnerability exists in PraisonAI versions 4.6.34 through 4.6.57, a multi-agent teams system. The application accepts an --api-key parameter via the 'praisonai serve agents' command, but the internal _create_agents_app() function fails to enforce authentication on POST /agents and POST /agents/{agent_name} endpoints. This allows any network caller to invoke configured agents without supplying credentials, even when an API key has been configured. The vulnerability essentially renders the API key mechanism ineffective, exposing agent functionality to unauthorized access. The flaw was fixed in version 4.6.58 via a committed patch. Users are advised to upgrade immediately to the patched release to prevent unauthorized agent invocation.
Bekijk origineel advisory →Nokogiri versions prior to 1.18.4 bundle a vulnerable version of libxslt (prior to 1.1.43) that contains two use-after-free vulnerabilities. CVE-2025-24855 involves a use-after-free of the XPath context node due to xsltEvalXPathStringNs leaking xpathCtxt->node. CVE-2024-55549 involves a use-after-free related to excluded result prefixes and namespaces. Processing specially crafted XSLT input can trigger memory corruption exploiting these flaws. The vulnerabilities are inherited through Nokogiri's bundled dependency on libxslt, representing a supply chain risk. Nokogiri 1.18.4 resolves both issues by upgrading the bundled libxslt to version 1.1.43. Users of Nokogiri are advised to upgrade to version 1.18.4 or later immediately.
Bekijk origineel advisory →NLTK versions before 3.10.3 contain a vulnerability in the dependencygraph.dot2img and AlignedSent._repr_svg_ functions that fail to use validated absolute paths when invoking the Graphviz dot binary. This allows attackers to execute arbitrary code by placing a malicious dot binary in the search path or current working directory. On Windows systems, attackers can exploit bare-name binary resolution via the current working directory, while on Unix-like systems they can abuse relative PATH entries. The vulnerability enables a classic binary hijacking/path traversal attack vector. Successful exploitation results in full arbitrary code execution with the privileges of the NLTK process. Users are advised to upgrade to NLTK version 3.10.3 or later to remediate the issue.
Bekijk origineel advisory →A high-severity vulnerability (CVE-2026-65083) has been identified in NVIDIA OpenShell for Linux, specifically within its sandbox provisioning API. The flaw allows an attacker to cause an incomplete list of disallowed inputs, effectively bypassing sandbox restrictions. Successful exploitation of this vulnerability can lead to multiple severe impacts including arbitrary code execution, escalation of privileges, information disclosure, data tampering, and denial of service. The vulnerability is classified with a high criticality rating. NVIDIA has published a security advisory through their product-security GitHub repository. The issue affects the Linux version of NVIDIA OpenShell and is tracked by both the NVD and CVE databases. Users and administrators running affected versions are advised to review NVIDIA's advisory for remediation guidance.
Bekijk origineel advisory →CVE-2026-75037 describes an authentication bypass vulnerability in LACT (Linux AMD GPU Control Tool) affecting versions through 0.10.0. The flaw resides in the Polkit authentication mechanism, which relies on UnixProcessSubject and Peer PID for identity verification. This approach is susceptible to PID reuse attacks, allowing an attacker to bypass authentication checks. The vulnerability runs on Linux systems where LACT is installed and used for GPU management. A fix has been committed via commit d0478fe42c2219454e272f96b1cbd29ab37ee566 in the upstream GitHub repository. The issue was also tracked in the SUSE Bugzilla system under bug ID 1276480. Users are advised to update to a patched version beyond 0.10.0 to mitigate this risk.
Bekijk origineel advisory →Nokogiri versions before 1.15.6 and 1.16.x before 1.16.2 (CRuby with packaged libxml2) are affected by a use-after-free vulnerability in libxml2 tracked as CVE-2024-25062. The flaw exists in the xmlTextReader module, which underlies Nokogiri::XML::Reader. Exploitation requires using the XML Reader interface with DTD validation and XInclude expansion enabled, where processing a crafted XML document can trigger an xmlValidatePopElement use-after-free condition. The vulnerability is resolved in Nokogiri 1.15.6 and 1.16.2 by upgrading the packaged libxml2 to versions 2.11.7 and 2.12.5 respectively. JRuby-based installations and those using the system-installed libxml2 are not affected by this issue.
Bekijk origineel advisory →CVE-2026-45019 affects Chainlit, a Python framework for conversational AI, in versions 2.4.0rc0 through 2.12.0. When the MCP feature is enabled, the POST /mcp endpoint is exposed without authentication. Attackers can supply arbitrary URLs and headers to SSE and streamable-HTTP transports, causing the Chainlit server to make blind outbound requests to internal or external services. This enables Server-Side Request Forgery (SSRF) attacks, including probing cloud metadata endpoints, internal port scanning, and issuing state-changing authenticated requests using attacker-controlled Authorization and Cookie headers. The SSE sink has been present since version 2.4.0rc0, with header forwarding and streamable-HTTP support added in 2.6.4. The vulnerability is caused by missing scheme validation, private-address filtering, and allowlist enforcement in backend/chainlit/types.py and server.py. The issue is fully remediated in Chainlit version 2.12.0.
Bekijk origineel advisory →A Server-Side Request Forgery (SSRF) vulnerability exists in PraisonAI versions prior to 4.6.58. The flaw resides in JobSubmitRequest.validate_webhook_url(), which incorrectly accepts webhook URLs when DNS resolution fails with a socket.gaierror exception due to a silent exception handler (except socket.gaierror: pass). A subsequent fresh DNS lookup in JobExecutor._send_webhook() can be exploited via DNS rebinding, allowing an attacker to redirect requests to internal services. This constitutes a classic Time-of-Check to Time-of-Use (TOCTOU) vulnerability in DNS validation. The issue enables attackers to potentially access internal infrastructure not intended to be publicly reachable. The vulnerability has been patched in PraisonAI version 4.6.58. A GitHub security advisory (GHSA-rg5q-pp8p-f7jm) has been published alongside the fix. Users are strongly advised to upgrade to the patched version immediately.
Bekijk origineel advisory →CVE-2026-24170 affects NVIDIA UFM Enterprise, specifically its web interface authorization component. An authenticated user can exploit this vulnerability by sending specially crafted HTTP requests, triggering improper authentication behavior. Successful exploitation may lead to arbitrary code execution and escalation of privileges on the affected system. The vulnerability requires an authenticated session to exploit, slightly limiting the attack surface but not eliminating the risk. NVIDIA has published an advisory through their product-security GitHub repository. The flaw is catalogued by both NVD and CVE.org. Given the potential for code execution and privilege escalation, the vulnerability is rated high criticality. Organizations using NVIDIA UFM Enterprise should review the vendor advisory and apply patches promptly.
Bekijk origineel advisory →Grav CMS versions before 2.0.16 contain a path traversal vulnerability in the MediaUploadTrait::deleteFile() method. Authenticated users with media management permissions can exploit this flaw to delete arbitrary files on the server by supplying filenames containing directory traversal sequences (../). The vulnerability exists because the method only validates the basename portion of the filename, while leaving the directory path portion unvalidated. These unvalidated paths containing ../ sequences are then passed directly to PHP's unlink() function, allowing deletion of files outside the intended media storage directory. The flaw is fixed in Grav version 2.0.16. This is a significant security risk as it can lead to deletion of critical system or application files, potentially causing denial of service or aiding further compromise.
Bekijk origineel advisory →CVE-2026-59981 affects the OpenEXR library, the reference implementation for the EXR image format used in the motion picture industry. The vulnerability exists in the OpenEXRUtil library's SampleCountChannel::row() API, which returns an out-of-bounds pointer when processing deep images with a non-zero dataWindow origin. The conflict between 0-based row indexing and absolute pixel coordinate offsets causes row(0) to point far outside the allocated sample-count buffer when the data window has a large negative vertical origin. An attacker can craft a malicious deep EXR file that, when opened by a vulnerable application, triggers an out-of-bounds read. This can result in process crashes or, under a controlled heap layout, exposure of adjacent heap memory contents. Affected versions include through 3.2.10, 3.3.0-3.3.12, and 3.4.0-3.4.13. Fixes are available in versions 3.2.11, 3.3.13, and 3.4.14.
Bekijk origineel advisory →CVE-2026-65096 is a high-severity vulnerability in NVIDIA NemoClaw for Linux affecting the Telegram bridge component. The flaw allows an attacker to perform OS command injection, potentially leading to arbitrary code execution on the affected system. Successful exploitation could also result in privilege escalation, information disclosure, and data tampering. The vulnerability is specific to the Linux version of the product. NVIDIA has published a security advisory referencing this CVE. The issue is tracked in NVIDIA's product security repository under the 2026/5872 path. No active exploitation details are currently mentioned, but the potential impact is significant given the range of consequences described.
Bekijk origineel advisory →CVE-2026-55546 affects QWED-MCP, a deterministic verification gateway for MCP, in versions prior to 0.2.1. The vulnerability exists in the verify_math_expression() function within math_engine.py, which passes attacker-controlled strings directly to SymPy's parse_expr() without adequate sanitization. Because parse_expr() internally calls Python's eval() with built-ins available, an attacker can leverage Python's import functionality to execute arbitrary OS commands. Potential impacts include arbitrary code execution, data exfiltration, secret theft, and access to internal services. Exploitation requires a downstream integration that passes untrusted user input to the library API, as the default MCP tool registry does not expose the vulnerable function. The issue has been patched in version 0.2.1 of QWED-MCP.
Bekijk origineel advisory →Ech0 versions 4.3.4 and earlier contain a vulnerability where scoped access token (least-privilege) restrictions are not reliably enforced on several privileged admin API routes. Multiple endpoints such as /api/inbox, /api/panel/comments, and /api/backup/export omit scope checks and authorize requests based solely on the user's admin role. The backup export handler additionally discards token scope metadata entirely. An attacker with a deliberately limited low-scope admin access token can access broader privileged functionality than intended, including reading the inbox and exporting a full database backup as a ZIP archive. This represents a broken access control / authorization bypass issue. The vulnerability has been fixed in version 4.4.3. The issue is documented in the NVD, GitHub Security Advisories, and VulnCheck advisories.
Bekijk origineel advisory →