Overview of incoming advisories.
1553 results found
CVE-2026-80049 describes a critical authorization bypass vulnerability in Airbyte Platform through version 2.0.0. The flaw exists because the platform resolves workspace authorization decisions based on a caller-supplied field rather than the actual resource's owning workspace. AuthorizationServerHandler extracts identifiers from raw JSON request bodies into X-Airbyte-* headers, and AuthenticationHeaderResolver.resolveWorkspace prioritizes the caller-supplied X-Airbyte-Workspace-Id over resource-derived headers. Endpoints that only expect a resource identifier in the request body can be reached with an injected workspaceId field, causing permission checks to run against the attacker-nominated workspace. Since no post-authorization check compares the resource's owning workspace with the authorized workspace, any authenticated workspace member can read source and destination configurations, trigger or cancel syncs, and delete connections, sources, and destinations belonging to other workspaces. The exploit requires only valid membership in any workspace, making it broadly exploitable across multi-tenant deployments.
View original advisory →NLTK versions up to and including 3.9.4 contain an unsafe pickle deserialization vulnerability in the TransitionParser.parse() method located in nltk/parse/transitionparser.py. The method invokes pickle_load() with the default parameter restricted=False, routing deserialization through WarningUnpickler, which does not override find_class() and thus permits arbitrary class resolution. An attacker can craft a malicious model file containing pickle gadget chains that execute arbitrary Python code with the privileges of the application user upon loading. While NLTK provides a RestrictedUnpickler class designed for safe deserialization, it is not utilized in production code paths, leaving users exposed. The vulnerability enables remote or local code execution depending on how model files are sourced and loaded. The issue has been addressed and fixed in NLTK version 3.10.0. Users are strongly advised to upgrade immediately to mitigate the risk of arbitrary code execution.
View original advisory →CVE-2026-55099 affects the Python icalendar library versions 7.1.0 through 7.1.2, where the Component equality method performs O(2^n) recursive comparisons on nested subcomponents. An attacker can craft a sub-kilobyte .ics file with deeply nested BEGIN:VEVENT blocks to trigger exponential CPU consumption. The vulnerability is exploited when applications perform equality, inequality, membership, deduplication, or normalization comparisons on parsed calendar data. Affected endpoints include calendar sync, import, invite processing, and any comparison paths. Parsing alone does not trigger the issue, but equal nested subtrees cause worst-case behavior. A few hundred bytes are sufficient to pin a CPU core for minutes or indefinitely, resulting in denial of service. The issue has been patched in icalendar version 7.1.3.
View original advisory →CVE-2026-65093 is a high-severity vulnerability in NVIDIA OpenShell for Linux that allows an attacker to perform a sandbox escape. Successful exploitation can lead to arbitrary code execution, escalation of privileges, data tampering, and information disclosure. The vulnerability was published via the NVD (National Vulnerability Database) and tracked by NVIDIA's product security team. No specific exploit details or active exploitation in the wild are mentioned in the article. The vulnerability affects Linux-based deployments of NVIDIA OpenShell. Given the potential for privilege escalation and code execution, this is classified as a high-criticality issue. Users and administrators running NVIDIA OpenShell on Linux systems should monitor NVIDIA's security advisories for patches and mitigations.
View original advisory →A vulnerability was identified in Fdawgs node-poppler versions up to 9.1.2 and 10.0.1 affecting multiple PDF processing functions in src/index.js. The flaw resides in the Argument Injection Handler component, where manipulation of the file_path argument leads to argument injection. Affected functions include pdfInfo, pdfToText, pdfToCairo, pdfToPpm, pdfImages, pdfToHtml, pdfToPs, pdfFonts, pdfDetach, pdfAttach, pdfSeparate, and pdfUnite. The vulnerability can be exploited remotely, increasing its risk surface. A patch has been identified with commit hash db6e3f79d3beb20601be7e59669c39811ae3c330 on the project's GitHub repository. Users are strongly advised to apply the patch immediately to mitigate the risk of exploitation.
View original advisory →CVE-2026-56710 affects Grav Login plugin versions before 1.0.16, where the onApiUserListRowAction unlock handler fails to validate the privilege level of the target account. An attacker holding api.users.write permission can exploit this flaw to clear login lockout counters on admin.super accounts. This effectively strips brute-force protection from the highest-privilege accounts in a Grav CMS installation without requiring equivalent super-admin permissions. The vulnerability represents a privilege escalation/authorization bypass issue that could facilitate subsequent credential-based attacks against administrator accounts. It has been assigned a high criticality rating. Fixes are available in version 1.0.16 of the Grav Login plugin, and advisories have been published on GitHub and VulnCheck.
View original advisory →CVE-2026-63072 is a vulnerability in OpenSSL's CMS decryption logic where the key-unwrap output buffer is sized based on the reported unwrapped key size, but the AES-WRAP-PAD primitive can write 8 bytes beyond the allocation. An attacker can craft a CMS message by changing a single OID byte to trigger a deterministic 8-byte out-of-bounds heap write when the victim calls CMS_decrypt(). This results in heap corruption, typically causing a Denial of Service. The vulnerability is classified as CWE-787 (Out-of-bounds Write), requires no special configuration, and is reachable via the public API. No FIPS modules are affected as the CMS code resides outside the FIPS module boundary. The fix sizes the unwrap output buffer for the worst-case scenario to prevent out-of-bounds writes. Multiple OpenSSL commits have been issued to address the issue.
View original advisory →CVE-2026-45018 affects Chainlit, a Python framework for conversational AI applications, in versions 2.4.0rc0 through 2.12.0. When the MCP feature is enabled, the POST /mcp endpoint is exposed without authentication. The validate_mcp_command() function only checks the executable name but fails to validate arguments passed to StdioServerParameters. An attacker can exploit npx's -c argument to execute arbitrary shell commands with Chainlit process privileges. If allowed_executables is unset, the None default permits any executable, worsening the exposure. The vulnerability requires no authentication to exploit, making it accessible to unauthenticated remote attackers. The issue has been patched in version 2.12.0. Organizations running Chainlit with MCP enabled should upgrade immediately.
View original advisory →Ech0 versions through 4.5.6 contain an OAuth redirect URI validation vulnerability in the parseAndValidateClientRedirect function (internal/service/auth/auth.go). The validation logic only compares the scheme and host of the client-supplied redirect_uri against an admin-configured allowlist, ignoring path, query, and fragment components. The unvalidated redirect_uri is embedded into a signed state JWT at login time. An attacker can craft a redirect_uri with a matching host but an attacker-controlled path; after the OAuth exchange, the victim is redirected to that path with a one-time exchange code in the query string. If the code leaks via Referer headers, analytics, or an open redirect on the allowed host, the attacker can redeem it at the public POST /api/auth/exchange endpoint to obtain the victim's access and refresh tokens, effectively achieving account takeover. The vulnerability has been fixed in version 4.7.3.
View original advisory →A critical authentication bypass vulnerability was discovered in PraisonAI, a multi-agent teams system, prior to version 4.6.58. The vulnerability exists in the create_auth_middleware() function, which fails to properly enforce authentication when the PRAISONAI_API_KEY or PRAISONAI_JWT_SECRET environment variables are not configured. As a result, an externally bound Recipe server can accept unauthenticated POST requests to /v1/recipes/run endpoints even when authentication is supposed to be enabled. This effectively renders the authentication mechanism non-functional under certain configuration conditions. The flaw could allow unauthorized users to execute recipes on exposed servers. The issue has been patched in PraisonAI version 4.6.58. Users are advised to upgrade immediately to mitigate the risk of unauthenticated access.
View original advisory →The Jawn theme for WordPress contains a critical Privilege Escalation vulnerability affecting all versions up to and including 1.4.2. Unauthenticated attackers can exploit this flaw to elevate their privileges to administrator level, posing a severe risk to affected WordPress installations. No authentication is required to exploit this vulnerability, making it particularly dangerous. The vulnerability has been documented by both Patchstack and Wordfence security researchers. WordPress site administrators using the Jawn theme should update to a patched version immediately. The CVE identifier assigned is CVE-2026-78477. The issue represents a significant access control failure within the theme's code. Sites running vulnerable versions are at risk of full administrative compromise.
View original advisory →NLTK versions prior to 3.10.3 contain a path sandbox bypass vulnerability affecting corpus-reader constructors. Attackers can supply arbitrary corpus root paths to LinThesaurusCorpusReader and PanLexLiteCorpusReader constructors, enabling unauthorized access to filesystem content and SQLite databases outside the pathsec sandbox boundary. The vulnerability allows reading of files outside the intended data root, posing a significant data exposure risk. The issue has been addressed in NLTK version 3.10.3. Advisories have been published on GitHub Security Advisories and VulnCheck. Users of affected versions should upgrade immediately to mitigate the risk of unauthorized file system access.
View original advisory →NLTK versions before 3.10.3 contain a vulnerability in the java() function that fails to validate JVM options passed via the per-call options parameter. This allows attackers to inject dangerous JVM flags such as -agentpath, -javaagent, or @argfile when interacting with Stanford wrapper classes. Successful exploitation can lead to arbitrary code execution on the affected system. The vulnerability is tracked as CVE-2026-79675 and has been assigned a high criticality rating. A fix is available in NLTK version 3.10.3 and later. Users of Stanford NLP wrapper classes within NLTK are particularly at risk. The issue was disclosed via GitHub Security Advisories and VulnCheck. Organizations using NLTK for natural language processing pipelines with Java-based Stanford tools should update immediately.
View original advisory →CVE-2026-55536 affects PraisonAI, a multi-agent teams system, in versions prior to 4.6.58. The vulnerability resides in the Browser Server's _handle_connection() function, which validates Chrome extension origins using Python's re.match() with an unanchored regular expression: chrome-extension://[a-z0-9]{32}. Because re.match() only matches from the start of the string without requiring a full match, extra trailing characters in the origin are not rejected, allowing the WebSocket connection to be accepted. This flaw enables attackers to send start_session commands and perform unauthorized browser automation. The vulnerability represents an improper input validation issue in the origin-checking logic. A fix was released in PraisonAI version 4.6.58, which addresses the regex anchoring problem. Users are advised to upgrade immediately to the patched version.
View original advisory →Adminer versions before 5.4.3 contain a critical remote code execution vulnerability in SQLite query handling. The vulnerability arises because the VACUUM INTO command is not blocked despite existing ATTACH restrictions. Authenticated attackers can exploit this flaw to write arbitrary PHP code to any file path on the server. Once malicious PHP files are written, attackers can execute arbitrary commands on the server. The vulnerability requires authentication to exploit, reducing but not eliminating its risk. A fix is available in Adminer version 5.4.3 and later. The issue has been documented in GitHub security advisories and VulnCheck advisories. Organizations using Adminer for database management should update immediately to mitigate this risk.
View original advisory →CVE-2026-59769 affects the Furuno FA-50 AIS transponder device, which contains hard-coded credentials in all versions. An attacker who knows these credentials and has access to the vessel's internal network can exploit them to access the settings screen. Through this access, the attacker can alter the vessel's identification number, posing a significant maritime safety and security risk. The vulnerability is classified under hard-coded credential weaknesses, a common but serious security flaw. No patch version is specified, suggesting all current firmware versions are affected. The issue was disclosed via JVN (Japan Vulnerability Notes) and Furuno's official security notices. This vulnerability is particularly concerning in maritime contexts where vessel identification integrity is critical for navigation safety and regulatory compliance.
View original advisory →A path traversal vulnerability has been identified in dekdee adobe-xd-mcp version 1.0.0. The flaw exists in the file src/parsers/xd-parser.ts within the file-access-from-request endpoint. Attackers can manipulate the outputFile or outputDir arguments to traverse the file system beyond intended directories. The vulnerability can be exploited remotely, increasing its risk surface significantly. A public exploit is already available, raising the threat level for any exposed instances. The project maintainer was notified via a GitHub issue but has not responded or issued a patch. No workaround has been officially provided at this time. The combination of public exploit availability and lack of vendor response makes this a high-priority concern.
View original advisory →CVE-2026-79786 affects Coroot versions 1.20.2 through 1.24.5, where the MCP OAuth dynamic client registration endpoint fails to validate redirect URIs. The endpoint is unauthenticated, allowing any attacker to register OAuth clients with redirect URIs pointing to attacker-controlled hosts. Attackers can craft and distribute malicious authorization URLs targeting signed-in users. When a victim approves consent, their authorization code is captured by the attacker. The attacker can then exchange the stolen authorization code for valid access tokens. This results in full MCP session hijacking. The vulnerability is a classic OAuth open redirect leading to authorization code interception. A fix is referenced in the Coroot GitHub repository at version 1.24.5's mcp_oauth.go file. Users of affected versions should upgrade immediately or restrict access to the registration endpoint.
View original advisory →CVE-2026-49845 is a SQL injection vulnerability in Apache Hive's Metastore component affecting all versions before 4.2.1. The flaw exists in the direct SQL partition-name resolution path, where client-supplied partition names are embedded into SQL queries via string concatenation rather than parameterized bind statements. Authenticated users with access to Hive Metastore APIs can craft malicious partition names containing single quotes to manipulate the generated WHERE clause. This allows attackers to read, modify, truncate, or affect unintended partition metadata within tables they can reference. The vulnerability is present when metastore.try.direct.sql is enabled, which is the default configuration. Impact is scoped to partition and table targeting rather than arbitrary cross-database access. The fix is available in Apache Hive version 4.2.1, which replaces string concatenation with proper parameterized queries. Users are strongly advised to upgrade immediately.
View original advisory →CVE-2026-55571 affects djust, a Phoenix LiveView-style reactive server-side rendering framework for Django with Rust-powered performance. Prior to version 1.0.4, the LiveViewConsumer.handle_mount method sends a navigate redirect frame when authentication or authorization denies a LiveView mount, but fails to close the WebSocket connection or clear the view instance. A malicious raw WebSocket client can ignore the redirect and retain the mounted socket. Since LiveViewConsumer.handle_event does not recheck authentication or authorization, attackers can send event frames to invoke event handler methods without a valid authenticated session. This flaw enables unauthorized sensitive data reads or mutations, including via handle_live_redirect_mount. The vulnerability has been patched in djust version 1.0.4.
View original advisory →