CVE Database

A use-after-free issue was addressed with improved memory management. This issue is fixed in Safari 26.6.1, iOS 18.7.10 and iPadOS 18.7.10, iOS 26.6.1 and iPadOS 26.6.1, macOS Tahoe 26.6.2, tvOS 27, visionOS 27, watchOS 27. Processing maliciously crafted web content may lead to an unexpected process termination.

An out-of-bounds access issue was addressed with improved bounds checking. This issue is fixed in Safari 26.6.1, iOS 18.7.10 and iPadOS 18.7.10, iOS 26.6.1 and iPadOS 26.6.1, macOS Tahoe 26.6.2, visionOS 27. Processing maliciously crafted web content may lead to an unexpected Safari crash.

The issue was addressed with improved input validation. This issue is fixed in Safari 26.6.1, iOS 18.7.10 and iPadOS 18.7.10, iOS 26.6.1 and iPadOS 26.6.1, macOS Tahoe 26.6.2, visionOS 27. Processing maliciously crafted web content may lead to an unexpected Safari crash.

The issue was addressed with improved checks. This issue is fixed in Safari 26.6.1, iOS 18.7.10 and iPadOS 18.7.10, iOS 26.6.1 and iPadOS 26.6.1, macOS Tahoe 26.6.2, visionOS 27. Processing maliciously crafted web content may lead to an unexpected Safari crash.

The issue was addressed with improved checks. This issue is fixed in Safari 26.6.1, iOS 18.7.10 and iPadOS 18.7.10, iOS 26.6.1 and iPadOS 26.6.1, macOS Tahoe 26.6.2, tvOS 27, visionOS 27, watchOS 27. Visiting a maliciously crafted website may leak sensitive data.

An information leakage was addressed with additional validation. This issue is fixed in iOS 18.7.10 and iPadOS 18.7.10, iOS 27 and iPadOS 27, macOS Golden Gate 27, tvOS 27, visionOS 27, watchOS 27. An app may be able to leak sensitive kernel state.

A use-after-free issue was addressed with improved memory management. This issue is fixed in Safari 26.6.1, iOS 18.7.10 and iPadOS 18.7.10, iOS 26.6.1 and iPadOS 26.6.1, macOS Tahoe 26.6.2, tvOS 27, visionOS 27, watchOS 27. Processing maliciously crafted web content may lead to an unexpected process crash.

Onyx is an open-source AI platform. Prior to 4.3.0, Onyx Enterprise Edition's PATCH /manage/admin/user-group/{user_group_id} and POST /manage/admin/user-group/{user_group_id}/add-users endpoints in ee/onyx/server/user_group/api.py call update_user_group and add_users_to_user_group in ee/onyx/db/user_group.py without enforcing _validate_curator_can_modify_group, allowing a curator to add accounts to arbitrary groups and obtain document access through get_acl_for_user and the OpenSearch access_control_list filter. This issue is fixed in version 4.3.0.

Dokploy is a free, self-hostable Platform as a Service (PaaS). Prior to 0.29.6, Dokploy's user.update procedure in apps/dokploy/server/api/routers/user.ts updates account.password without deleting other rows from session, allowing a compromised better-auth.session_token session to remain valid for up to three days after a password change. This issue is fixed in version 0.29.6.

The issue was addressed with improved memory handling. This issue is fixed in Safari 26.6.1, iOS 18.7.10 and iPadOS 18.7.10, iOS 26.6.1 and iPadOS 26.6.1, macOS Tahoe 26.6.2, visionOS 27. Processing maliciously crafted web content may lead to an unexpected Safari crash.

A reachable assertion was addressed with improved input validation. This issue is fixed in iOS 18.7.10 and iPadOS 18.7.10, iOS 26.5 and iPadOS 26.5, macOS Tahoe 26.5, visionOS 26.5, watchOS 26.5. An attacker in a privileged network position may be able to cause a denial-of-service.

The issue was addressed with improved memory handling. This issue is fixed in Safari 26.5, iOS 18.7.10 and iPadOS 18.7.10, iOS 26.5 and iPadOS 26.5, macOS Tahoe 26.5, tvOS 26.5, visionOS 26.5, watchOS 26.5. Processing maliciously crafted web content may lead to an unexpected Safari crash.

This vulnerability only affects Grafana stacks configured with multiple organizations; single-organization deployments are not impacted. In a multi-organization stack, a user who is an Org Admin of a single organization can call GET /api/access-control/users/permissions/search?actionPrefix=dashboards: and receive permission data belonging to other organizations. The disclosed data is limited to dashboard and folder identifiers (UIDs) and per-user permission/scope mappings (which user holds which access on which dashboard). Dashboard contents, panels, query results, datasource credentials, secrets, and personal data are not exposed. This is a limited cross-organization information disclosure affecting multi-org deployments only.

Mattermost versions 11.7.x <= 11.7.6, 10.11.x <= 10.11.21, 11.8.x <= 11.8.3 fails to validate WebSocket command field types which allows an authenticated user to crash the plugin process and deny service to all Boards users via a custom_focalboard_SUBSCRIBE_TEAM message with a non-string teamId.. Mattermost Advisory ID: MMSA-2026-00687

Pandora is affected by a stored cross-site scripting vulnerability in the PDF download functionality. The /task-download/<task_id>/.../pdf endpoint verifies that the submitted file is a PDF using Pandora's content-based file-type detection, but previously returned the file using send_file(task.file.path) without explicitly specifying the MIME type or forcing it to be downloaded as an attachment. Because Flask determines the response MIME type from the filename when a path is supplied, an attacker could submit a file whose content is recognized by Pandora as a PDF while its filename or extension causes the download endpoint to return it with a different, potentially active MIME type. A specially crafted PDF/polyglot file could therefore be served inline and interpreted by a victim's browser as HTML or another executable web format. If a victim with access to the submitted analysis follows the PDF download link, attacker-controlled script could execute in the security context of the Pandora application, potentially allowing access to application data or actions using the victim's session. The patch prevents the issue by explicitly returning PDF downloads with Content-Type: application/pdf, forcing Content-Disposition: attachment, and assigning a trusted .pdf filename based on the task UUID.

OpenViking debug vector scroll and count endpoints apply only account-level scoping without user-level access controls, allowing authenticated users to read all co-tenant records. Attackers can query these endpoints to retrieve private memories, resources, skills, and secret material belonging to other users in the same account without administrative privileges.

Next Terminal fails to enforce per-asset authorization checks on the portal ping and wake-on-LAN endpoints, allowing any authenticated user to probe and wake assets they are not granted access to. Attackers can call these endpoints with arbitrary asset identifiers to retrieve asset information including display names, reachability status, connection timing, and network addresses, or trigger wake-on-LAN packets on unauthorized assets.

Hugging Face Transformers fails to validate shard filenames in checkpoint index files, allowing attackers to read arbitrary files outside the model directory. Attackers can supply malicious index files with parent-directory references or absolute paths that are joined without validation, enabling file disclosure and filesystem reconnaissance.

vLLM is an inference and serving engine for large language models. Prior to 0.26.0, the MiMoV2OmniMultiModalProcessor in vllm/transformers_utils/processors/mimo_v2_omni.py passes attacker-controlled image and audio strings through _fetch_image, requests.get, and Image.open instead of MediaConnector, bypassing allowed_media_domains and allowed_local_media_path protections and allowing server-side requests and reads of arbitrary files accessible to the vLLM process. This issue is fixed in version 0.26.0.

hashcat master branch builds after v7.1.2 contain a heap buffer overflow vulnerability in the KeePass AESKDF/KDBX v4 module (module 34301) that allows attackers to corrupt adjacent heap memory by supplying an oversized ninth hash field token. The module accepts up to 600 hex characters for the ninth token field but decodes it into a fixed 256-byte buffer with no length check, allowing a maximal input to write up to 44 bytes past the buffer boundary into adjacent esalt fields and heap chunk metadata, potentially enabling heap corruption or memory access violations.

Cross Site Scripting vulnerability in JeecgBoot v.3.9.2 allows a remote attacker to execute arbitrary code via the endpoint /airag/chat/upload

Deskflow is a keyboard and mouse sharing app. From 1.17.0 until continuous build 1.26.0.300, a connected peer can send repeated DCLP DataChunk messages to ClipboardChunk::assemble() in src/lib/deskflow/ClipboardChunk.cpp, causing the server path in src/lib/server/ClientProxy1_6.cpp or client path in src/lib/client/ServerProxy.cpp to append data beyond the DataStart declared size and configured clipboard limit before DataEnd validation, exhausting receiver memory. This issue is fixed in continuous build 1.26.0.300.

JumpServer is an open source bastion host and an operation and maintenance security audit system. From 4.8.0 until 4.10.17, an authenticated user with SFTP permission to an authorized asset can submit crafted traversal paths through the KoKo Web Terminal SFTP feature, causing AssetDir.GetRealPath() in pkg/srvconn/sftp_asset.go to resolve paths outside the intended SFTP root and permit read, list, write, rename, or delete operations under the configured backend account on that asset. This issue is fixed in version 4.10.17.

JumpServer is an open source bastion host and an operation and maintenance security audit system. Prior to 4.10.17, a user with the users.invite_user permission can submit an existing member to POST /api/v1/users/users/invite/, causing the organization invitation logic in apps/users/api/user.py to execute user.org_roles.set(org_roles) and replace the member's existing organization roles, which can escalate privileges or downgrade administrators. This issue is fixed in version 4.10.17.

JumpServer is an open source bastion host and an operation and maintenance security audit system. Prior to 4.10.17, an authenticated administrator with Applet Host management and deployment permissions can inject Jinja2 expressions into the IP/Host field or Core Service Address field, causing Ansible to evaluate ansible_host inventory data or playbook variables during Applet Host deployment and execute arbitrary commands on the JumpServer control node. This issue is fixed in version 4.10.17.

OpenEMR before 8.2.0 contains a path traversal vulnerability in the standard_tables_manage.php interface where the db GET parameter is passed without validation to temp_dir_cleanup(), which joins the value to the PHP temporary directory path and recursively deletes the resulting directory. Attackers can supply a traversal sequence in the db parameter to resolve outside the intended temporary directory, and by chaining this with an open redirect in dicom_frame.php, an unauthenticated attacker can deliver a crafted URL that triggers arbitrary recursive directory deletion within an authenticated Superuser's session.

A vulnerability was detected in TOTOLINK EX1200L 9.3.5u.6146_B20201023. This affects the function setWizardCfg of the file /cgi-bin/cstecgi.cgi. The manipulation results in null pointer dereference. The attack can be launched remotely. The exploit is now public and may be used.

A security vulnerability has been detected in TOTOLINK EX1200L 9.3.5u.6146_B20201023. Affected by this issue is the function setPasswordCfg of the file /cgi-bin/cstecgi.cgi of the component Password Configuration Handler. The manipulation leads to null pointer dereference. The attack can be initiated remotely. The exploit has been disclosed publicly and may be used.

Notepad++ is a free and open-source source code editor. Prior to 8.9.7, macros loaded from an attacker-controlled shortcuts.xml bypass the HMAC validation applied to UserDefinedCommands and can invoke Scintilla actions and the internal Open in Default Viewer command in an elevated Notepad++ process, allowing protected file modification and conditional elevated command execution when a local attacker influences settingsDir and a user triggers the macro. This issue is fixed in version 8.9.7.

vLLM is an inference and serving engine for large language models. Prior to 0.26.0, the /v1/completions/derender and /v1/chat/completions/derender endpoints accept caller-supplied GenerateResponse objects whose generate_responses, choices, token_ids, prompt_logprobs, logprobs.content, top_logprobs, and routed_experts structures are processed by OnlineDerenderer and tokenizer.decode before max_model_len, max_tokens, max_num_seqs, or response-size limits are enforced, allowing an authenticated API client to consume excessive CPU and memory and produce oversized responses. This issue is fixed in version 0.26.0.