直近表示から外れた情報を確認できます。診断結果とは別情報として扱います。
Spring Statemachine's Kryo-based persistence backends (JPA, MongoDB, Redis and ZooKeeper) deserialise persisted state-machine contexts without enforcing a class allowlist (CWE-502, deserialisation of untrusted data), which can lead to remote code execution inside the application JVM. Affected versions: Spring Statemachine 4.0.0 through 4.0.1 Spring Statemachine 3.2.0 through 3.2.4
FOSSBilling is a free, open-source billing and client management system. In versions 0.7.2 and prior, a query-construction flaw in client list endpoints allowed authenticated clients to bypass tenant scoping and retrieve other clients’ data. Details In ServiceTransaction::getSearchQuery() and Order\Service::getSearchQuery(), OR-based search/action filters were appended without grouping, allowing SQL operator precedence to evaluate OR clauses independently of the enforced client_id constraint. Crafted requests could therefore return records and metadata belonging to other clients, including identifiers, amounts, status, timestamps, and related fields. This issue was fixed in version 0.8.0.
A flaw was found in GStreamer's gst-plugins-bad package. When processing a specially crafted H.264 video file containing malformed MVC or SVC extension slice NAL units, a 1-byte heap out-of-bounds read can occur during parsing. This happens when the parser attempts to check slice boundary information without first verifying that the NAL unit contains enough data beyond the extension header. An attacker could exploit this by tricking a user into opening a malicious H.264 video file, potentially causing the application to crash or leak a single byte of heap memory.
A flaw was found in the GStreamer gst-plugins-bad package. When processing a malformed H.266/VVC video stream with a crafted aspect ratio indicator value, the H.266 parser performs an out-of-bounds read of up to 8 bytes from adjacent memory. This flaw allows an attacker to craft a malicious H.266 video file or stream that, when processed by a GStreamer-based application, could leak limited memory contents through video metadata, potentially exposing sensitive information from the application's address space.
A flaw was found in the foreman-mcp-server. A session management vulnerability in the MCP Server allows unauthenticated attackers to hijack active administrative sessions due to an improper cache of authenticated client connections, by trusting a non-secret session ID without re-validating authentication tokens and by logging all newly created session IDs to standard logs. This issue can result in privilege escalation and infrastructure-wide code execution.
A flaw was found in the community.general Ansible collection's nexmo module. The module constructs HTTP requests to the Vonage/Nexmo SMS API by encoding API credentials (api_key and api_secret) into URL query parameters and sending them via GET requests. This causes credentials to be exposed in web server access logs, proxy logs, HTTP Referer headers, and network monitoring tools, despite the Ansible argument specification marking these parameters as no_log. An attacker with access to any of these logging or monitoring points can obtain the full API credentials and gain unauthorized access to the victim's Vonage/Nexmo account.
Module: plugins/modules/keyring_info.py CVSS 3.1: 5.5 MEDIUM — AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:N Issue: The module retrieves a passphrase from the OS native keyring (GNOME Keyring, macOS Keychain, Windows Credential Manager) and places it directly into result["passphrase"] with no output suppression, no no_log protection, and no documentation warning. Root Cause: Line 105 (protected): keyring_password=dict(type="str", required=True, no_log=True) Line 127 (NOT protected): result["passphrase"] = passphrase Observed Output: { "changed": false, "passphrase": "MyMasterP@ssw0rd!SSH_Key_Secret" } Visible via register + debug: { "keyring_result": { "changed": false, "passphrase": "MyMasterP@ssw0rd!SSH_Key_Secret" } } Impact: Master passwords, SSH key passphrases and service credentials appear in all Ansible output register: keyring_result followed by debug: var=keyring_result prints passphrase in full Ansible fact caching backends (Redis, JSON file, memcached) may persist the passphrase AWX/Tower job logs silently store the live credential Fix: module.exit_json(changed=False, passphrase=passphrase, _ansible_no_log=True) Also add a documentation warning requiring callers to use no_log: true at the task level. PoCs Fig 1: PoC execution showing passphrase in plaintext output Fig 2: Source code showing no_log=True on input (line 105) vs unprotected output (line 127)
A missing authorization vulnerability was found in the Event-Driven Ansible (EDA) websocket API. The /api/eda/ws/ansible-rulebook endpoint does not verify user permissions when processing Worker messages. Any authenticated user can send a forged message with an arbitrary activation_id to receive plaintext credentials associated with that activation, including OAuth tokens, vault passwords, and SSH keys.
FOSSBilling is a billing and client management system that automates invoicing, payments, and communication for online service businesses. Versions 0.6.21 through 0.7.2 are vulnerable to IDOR through the support ticket creation workflow. By manipulating rel_id when rel_type=order, an authenticated client can create a support ticket that references another client's order they do not own. The ticketCreateForClient() method accepted rel_id without verifying order ownership for non-upgrade tasks, allowing clients to link a new ticket to another client's order by crafting the request. No cron task automatically processes cancel/upgrade requests from ticket relations; staff action is required. This affects integrity and confidentiality: staff could be misled into acting on the wrong order (e.g., cancellation or upgrade requests). While there is no client-to-client order data exposure, order IDs may appear in ticket context. This issue has been fixed in version 0.8.0.
Traefik is an HTTP reverse proxy and load balancer. From 3.7.0-ea.1 until 3.7.5, there is a medium severity vulnerability in Traefik's Kubernetes Ingress NGINX provider that causes affected routes to fail open. When an Ingress explicitly enables BasicAuth or DigestAuth through the supported nginx.ingress.kubernetes.io/auth-type and auth-secret annotations, but the referenced auth Secret cannot be resolved or parsed, Traefik logs the resolution error, skips installing the authentication middleware, and still emits a router to the backend service. A route that operators intended to protect is therefore published to the data plane without its authentication control, allowing unauthenticated access to the backend. The trigger is an invalid or unresolved auth dependency — a missing, malformed, unreadable, or policy-denied Secret — rather than an intentionally unprotected route. This vulnerability is fixed in 3.7.5.
Traefik is an HTTP reverse proxy and load balancer. Prior to 3.6.21 and 3.7.5, there is a high severity vulnerability in Traefik's Kubernetes Gateway provider affecting the crossProviderNamespaces allowlist. For HTTPRoute rules that declare multiple (WRR) backendRefs, Traefik evaluates the allowlist against the target backendRef.namespace instead of the route's own namespace. As a result, an HTTPRoute created in a namespace that is not allow-listed can reference a cross-provider TraefikService such as api@internal, dashboard@internal or rest@internal by pointing backendRef.namespace at an allow-listed namespace covered by a Gateway API ReferenceGrant, exposing internal Traefik services on the data plane. Exploitation requires the ability to create an accepted HTTPRoute and a matching ReferenceGrant from an allow-listed namespace; it does not require any change to Traefik static configuration, RBAC, or the deployment itself. This vulnerability is fixed in 3.6.21 and 3.7.5.
rtk filters and compresses command outputs before they reach your LLM context. Prior to 0.42.2, the permission splitter did not conservatively split or reject several shell constructs that Bash treats as command execution boundaries or nested execution. As a result, a command beginning with an allowed prefix such as git could hide a second command behind one of these constructs. rtk rewrite returned exit code 0, causing the Claude hook to emit permissionDecision: "allow". The rewritten command still contained the hidden command, so it ran without the user confirmation or denial that the permission rules were intended to enforce. This vulnerability is fixed in 0.42.2.
Pi is a minimal terminal coding harness. From 0.74.0 until 0.78.1, Pi versions with temporary npm or git extension package installs used predictable paths under the operating system temporary directory. On Linux-based multi-user systems, a local attacker who can write to the shared temporary directory could prepare the expected package location before another user runs pi with a temporary extension package source. Pi could then load attacker-controlled extension code in the victim user's process. This vulnerability is fixed in 0.78.1.
Pi is a minimal terminal coding harness. From 0.74.0 until 0.78.1, Pi stored API keys and OAuth credentials in auth.json. A race condition in the file write path could briefly create or rewrite this file with permissions derived from the process umask before tightening the file to owner-only permissions. This vulnerability is fixed in 0.78.1.
Pi is a minimal terminal coding harness. From 0.74.0 until 0.78.1, Pi HTML exports render session Markdown into a static HTML file. It did not consistently reject unsafe Markdown link and image URL schemes. In versions with scheme filtering, C0 control characters in the URL scheme could bypass the check because browsers normalize those characters before navigation. This vulnerability is fixed in 0.78.1.
Pi is a minimal terminal coding harness. Pi before 0.79.0 loaded project-local configuration and resources from a repository's .pi directory without first asking the user to trust that repository. This included project-local extensions, which are executable TypeScript or JavaScript modules loaded into the Pi process. An attacker who controls a repository could place Pi-specific project resources in that repository. If a user then started Pi from that working tree, the project-local extension code could run with the same privileges as the local Pi process without the user having a convenient way to make a trust decision. This vulnerability is fixed in 0.79.0.
Traefik is an HTTP reverse proxy and load balancer. Prior to 3.7.3, there is a critical vulnerability in Traefik's HTTP/3 (QUIC) TLS configuration selection that allows unauthenticated clients to bypass router-specific mTLS enforcement. When HTTP/3 is enabled on an entrypoint, the TLS handshake selects the applicable TLS configuration through an exact, case-sensitive lookup on the SNI value, which fails to match wildcard host patterns (e.g., *.example.com) or case variants of the configured hostname. Because the handshake falls back to the default TLS configuration — which may not require client certificates — a client can complete the QUIC handshake without presenting a certificate, while the subsequent HTTP routing layer still dispatches the request to a backend protected by a router-specific mTLS policy. The issue affects deployments where HTTP/3 is enabled, a router uses a wildcard Host rule or case-insensitive hostname matching, a router-specific TLSOptions enforces client certificate authentication, and UDP access to the entrypoint is reachable by an attacker. This vulnerability is fixed in 3.7.3.
Traefik is an HTTP reverse proxy and load balancer. From 3.7.0 until 3.7.3, there is a high severity vulnerability in Traefik's domain-fronting protection (SNICheck) that allows an unauthenticated client to bypass mutual TLS enforced through wildcard router TLSOptions. When a router uses a wildcard host rule such as Host(*.example.com) with stricter TLS options (for example RequireAndVerifyClientCert), SNICheck resolves the TLS options for the HTTP Host header using exact map lookups only and never applies wildcard matching. If another permissive SNI is served on the same entrypoint, an attacker can complete the TLS handshake under the permissive options and then send an HTTP Host header targeting the wildcard-protected backend, reaching it without presenting a client certificate. This affects the regular HTTPS / HTTP-2 path and does not require HTTP/3. This vulnerability is fixed in 3.7.3.
Traefik is an HTTP reverse proxy and load balancer. Prior to 2.11.48, 3.6.19, and 3.7.3, there is a high severity vulnerability in Traefik's StripPrefix middleware that allows an unauthenticated attacker to bypass route-level authentication and authorization. When a public router matches on a PathPrefix rule and applies the StripPrefix middleware, a request path containing .. or its percent-encoded form %2e%2e can match the public route at routing time and then, after the prefix is stripped and the path is normalized, resolve to a path served by a separate, authenticated router. As a result, an attacker can reach protected backend paths — such as admin or internal configuration endpoints — without satisfying the authentication middleware attached to the protected router. This vulnerability is fixed in 2.11.48, 3.6.19, and 3.7.3.
rtk filters and compresses command outputs before they reach your LLM context. Prior to 0.32.0, RTK (Rust Token Killer) improperly trusts project-local configuration files. RTK automatically loads .rtk/filters.toml from the working directory with highest priority and without user notification. An attacker can place a malicious filter file in a repository to apply regex-based modifications (e.g., strip_lines_matching) to shell command output before it is shown to the LLM, without any indication that the output has been modified. This allows attackers to selectively suppress or alter command output (including file contents, diffs, and security scan results) without detection, potentially concealing malicious code during AI-assisted development or review. This vulnerability is fixed in 0.32.0.
An issue in Pivotal CRM v.6.6.04.08 allows a remote attacker to execute arbitrary code via the Pivotal.Core.Common.dll and Pivotal.Engine.Client.Services.Conversion.dll components.
Improperly Controlled Modification of Dynamically-Determined Object Attributes vulnerability in ash-project ash allows a user to set the value of a private action argument that is intended to be controlled only by trusted server-side code. Action arguments declared with public?: false are meant to be set internally (for example via Ash.Changeset.set_private_argument/3) and must not be settable from end-user input. When a changeset is built from a parameter map, Ash filters out private arguments, but the filtering is incomplete. In the regular changeset path (for_create, for_update, for_destroy), private arguments are stripped only when the parameter key is an atom. When the key is a binary (string), as is the case for user-supplied parameters, the private argument is kept and the user controls its value. In the atomic path (Ash.Changeset.fully_atomic_changeset/4, also reached through atomic and bulk updates), private arguments are not stripped at all, regardless of whether the key is an atom or a binary. An attacker who can submit parameters to an action that defines a private argument can therefore inject a value for that argument. Depending on how the application uses the argument (for example an acting_user_id driving authorization or record ownership), this can lead to an integrity violation or privilege escalation. This issue affects ash: from 3.0.0 before 3.29.3.
@rtk-ai/rtk-rewrite transparently rewrites shell commands executed via OpenClaw's exec tool to their RTK equivalents. In 1.0.0, the @rtk-ai/rtk-rewrite OpenClaw plugin passes attacker-controlled input directly into a shell-backed execSync() template string without shell-safe escaping. JSON.stringify() wraps the value in double quotes and escapes inner double-quotes and backslashes, but leaves $() and backtick shell metacharacters untouched. Because execSync delegates execution to /bin/sh -c, the shell expands $(...) substitutions even inside double-quoted strings, causing the injected subcommand to execute before rtk is invoked. An attacker who can influence the exec tool's command parameter (e.g., via an LLM agent prompt or gateway/tool-call input) achieves arbitrary OS command execution with the privileges of the plugin/gateway process.
Daytona is a secure and elastic infrastructure runtime for AI-generated code execution and agent workflows. Prior to 0.185.0, Daytona's organization role update and delete endpoints authorized the caller as an owner of the organization named in the request path, but resolved and mutated the target role by its identifier alone, without verifying the role belonged to that organization. An authenticated user who owns any organization (organizations are self-service) could therefore modify the permissions of, or delete, a role belonging to a different organization, given that role's identifier. This vulnerability is fixed in 0.185.0.
Daytona is a secure and elastic infrastructure runtime for AI-generated code execution and agent workflows. From 0.101.0 until 0.184.0, sandbox previews that were switched from public to private could remain reachable without authentication for a short period after the change, due to a cached visibility state that was not invalidated when the sandbox's visibility changed. This vulnerability is fixed in 0.184.0.
Daytona is a secure and elastic infrastructure runtime for AI-generated code execution and agent workflows. Prior to 0.184.0, organization invitations could be accepted (and declined) by a user whose email matched the invitation but had not been verified. Daytona authenticates users via OIDC and matches an invitation's target email against the email in the caller's token, but the invitation accept and decline paths did not require that email to be verified, unlike organization creation, which already enforced verification. On identity providers that allow self-service signup and issue a session before the email is verified, an actor could register an address matching a pending invitation, leave it unverified, and accept the invitation, joining the target organization with the role the invitation carried (up to Owner). This vulnerability is fixed in 0.184.0.
Daytona is a secure and elastic infrastructure runtime for AI-generated code execution and agent workflows. Prior to 0.186, a sandbox volume reference (volumeId, which may also be a volume name) was forwarded to the runner and used to build the host bind-mount source path without confinement. A reference containing path-traversal sequences could in principle resolve the mount source outside the intended per-volume base directory. This vulnerability is fixed in 0.186.
Crawl4AI is an open-source LLM friendly web crawler & scraper. Prior to 0.8.9, the Docker API server applied its SSRF destination check to the crawl target URL only, not to the proxy address. An unauthenticated request could supply a proxy pointing at an internal IP and route the browser through it, reaching internal services and cloud-metadata endpoints, while using a perfectly valid crawl URL. The Docker API is unauthenticated by default. /crawl, /crawl/stream, and /crawl/job accept a browser_config (and crawler_config). The following all feed Chromium's egress and were unchecked: browser_config.proxy_config.server, browser_config.proxy (deprecated field), crawler_config.proxy_config.server, and --proxy-server / --proxy-pac-url / --proxy-bypass-list / --host-resolver-rules flags in browser_config.extra_args. This vulnerability is fixed in 0.8.9.
Crawl4AI is an open-source LLM friendly web crawler & scraper. Prior to 0.8.8, the Docker API server's SSRF protection (validate_webhook_url / validate_url_destination in deploy/docker/utils.py) used an explicit IPv4/IPv6 CIDR blocklist that missed several address families. An attacker could reach internal services and cloud metadata endpoints (e.g. 169.254.169.254) despite the filter by encoding an internal IPv4 address inside an IPv6 transition form, or by using the IPv6 unspecified address. Because the Docker API is unauthenticated by default (jwt_enabled: false), no credentials are required. This vulnerability is fixed in 0.8.8.
Crawl4AI is an open-source LLM friendly web crawler & scraper. Prior to 0.8.7, the _safe_eval_expression() function in the computed fields feature uses an AST validator that only blocks attributes starting with underscore. Python generator and frame object attributes (gi_frame, f_back, f_builtins) do NOT start with underscore, enabling a complete sandbox escape to achieve arbitrary code execution. The attack requires no authentication (JWT disabled by default) and is triggered via POST /crawl with a crafted extraction schema. This vulnerability is fixed in 0.8.7.
CMS (Cryptographic Message Syntax) parsing in gpgsm in GnuPG through 2.5.20 mishandles the CMS format for AES-GCM because aes-ICVlen is supposed to be 12 bytes but 4 bytes is accepted. NOTE: this is related to CVE-2026-34182.
GNU libidn before 1.44 is prone to out-of-bounds reads of uninitialized memory in the ToUnicode APIs because of mishandling in idna_to_unicode_internal. The affected code is not present in libidn2.
Deno is a JavaScript, TypeScript, and WebAssembly runtime. Prior to 2.7.5, a Deno program that opens a client WebSocket connection could be crashed by the remote server. While handling the WebSocket handshake response, Deno parsed the Sec-WebSocket-Protocol and Sec-WebSocket-Extensions response headers in a way that assumed their bytes were always printable ASCII. A response header containing non-visible-ASCII bytes (0x80-0xFF) caused a panic that aborted the entire Deno process. This vulnerability is fixed in 2.7.5.
Daytona is a secure and elastic infrastructure runtime for AI-generated code execution and agent workflows. Prior to 0.185.0, a cross-tenant authorization flaw in Daytona's notification WebSocket gateway allowed any authenticated user to subscribe to another organization's realtime notification channel and passively receive that organization's events. This vulnerability is fixed in 0.185.0.
Daytona is a secure and elastic infrastructure runtime for AI-generated code execution and agent workflows. Prior to 0.185.0, the daemon's git clone implementation disabled TLS certificate verification. When a clone request carried Git credentials, the daemon sent the HTTP Basic Authorization header to the remote over a connection whose certificate was never validated, on both the go-git and native git CLI code paths. An attacker able to intercept clone traffic could present any TLS certificate, capture the Git credentials supplied for the clone, and serve tampered repository content into the sandbox. This vulnerability is fixed in 0.185.0.
Home Assistant is open source home automation software that puts local control and privacy first. Prior to 2026.5.3, the LocationSensorManager BroadcastReceiver is exported with no permission. Any installed app, with zero runtime permissions, can broadcast a forged Google Play Services LocationResult directly to it; the receiver trusts the extra and forwards it to the user's Home Assistant server as the device's real location. This bypasses Android's developer-mode "Mock Location" gate and allows a local malicious app to drive zone-based automations (unlock door / disarm alarm / open garage) by faking the user's GPS position. This vulnerability is fixed in 2026.5.3.
Home Assistant is open source home automation software that puts local control and privacy first. Prior to 2026.6.0, the Konnected integration registers an HTTP endpoint, KonnectedView (homeassistant/components/konnected/__init__.py), that is marked as not requiring authentication (requires_auth = False). A comment next to that line says auth is instead handled "via the access token from configuration." That promise is only half true. Write requests (POST and PUT) are handled by update_sensor(), which does check the request's Authorization: Bearer <token> header against the integration's stored access tokens (using hmac.compare_digest). Read requests (GET) are handled by a separate get() method that has no authentication check at all. This vulnerability is fixed in 2026.6.0.
Claude Code is an agentic coding tool. From 0.2.54 until 2.1.163, because the hostname huggingface.co was pre-approved as a bare hostname for the WebFetch tool, any path on that domain—including attacker-controlled model repositories—was auto-approved without a permission prompt or being subject to --allowedTools restrictions. An attacker able to inject untrusted content into a Claude Code context could direct it to issue WebFetch requests against attacker-controlled repository files (e.g. /resolve/main/config.json), which HuggingFace counts as downloads server-side, creating a covert out-of-band channel for encoding and exfiltrating data Claude can access such as files, environment variables, or command output. Reliably exploiting this required the ability to add untrusted content into a Claude Code context window. This vulnerability is fixed in 2.1.163.
Electron is a framework for writing cross-platform desktop applications using JavaScript, HTML and CSS. From 42.3.1 until 42.3.3, Buffer performs incorrect byte length calculations resulting in heap buffer under/overflow. Most apps will crash and some may perform incorrect buffer allocations in the Node.js Buffer API resulting in unexpected truncation or allocation. This vulnerability is fixed in 42.3.3.
LobeHub is a work-and-lifestyle space to find, build, and collaborate with agent teammates that grow with you. Prior to 2.1.57, the /webapi/proxy endpoint on app.lobehub.com accepts a URL in the POST body and fetches it server-side without any authentication. An attacker can use this to make arbitrary outbound requests from LobeHub's infrastructure, leak Vercel deployment details, and inject cookies on the lobehub.com domain through reflected Set-Cookie headers. This vulnerability is fixed in 2.1.57.
Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to 0.8.11, the ydoc:document:join Socket.IO handler checks note ownership only when the document_id starts with note: (colon). However, the YdocManager storage layer normalizes all document IDs by replacing colons with underscores (document_id.replace(":", "_")). An attacker can join a document room using note_<id> (underscore) instead of note:<id> (colon), bypassing the authorization check entirely while accessing the same underlying Yjs document. The server then returns the full document state, leaking the victim's private note contents. This vulnerability is fixed in 0.8.11.
Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to 0.9.6, several direct, index-addressed Ollama proxy routes accept a caller-supplied url_idx path parameter and use it as a raw index into the admin-configured OLLAMA_BASE_URLS list. Access control on these routes validates only whether the user may use the requested model, never which backend the request is routed to. Any authenticated user can append an arbitrary url_idx to force their request onto an Ollama backend they were never authorized to reach, including internal, higher-privilege, or explicitly admin-disabled backends. This vulnerability is fixed in 0.9.6.
Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to 0.9.6, Open WebUI added collection-level ACL checks, but the patch can still be bypassed when Milvus multitenancy mode is enabled. The ACL allows unknown non-KB collection names as legacy/ephemeral collections. In Milvus multitenancy mode, that user-controlled collection name becomes a resource_id and is interpolated into a Milvus expression without escaping. This is caused by an incomplete fix for CVE-2026-44560 This vulnerability is fixed in 0.9.6.
Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to 0.9.6, the SafePlaywrightURLLoader implements a validate_url function to prevent SSRF attacks by checking the IP address of the user-provided URL. However, this validation is performed only on the initial URL. Since Playwright automatically follows HTTP redirects (301/302) by default, an attacker can bypass the validation by providing a safe URL that redirects to a restricted internal network address (e.g., localhost, Docker container network, or Cloud Metadata). This allows the application to access internal services despite ENABLE_RAG_LOCAL_WEB_FETCH being set to False This vulnerability is fixed in 0.9.6.
Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to 0.9.6, Open WebUI has a Broken Object Level Authorization (BOLA) vulnerability in the builtin search_knowledge_files tool. When native function calling is enabled and the selected model has no attached knowledge bases, an authenticated user can call search_knowledge_files with an arbitrary knowledge_id. The function then returns file metadata from that knowledge base without checking whether the user has read access. This allows unauthorized enumeration of private or restricted knowledge base files. This vulnerability is fixed in 0.9.6.
Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to 0.9.6, Open WebUI's prompt version-history endpoints authorize the prompt_id in the URL but then act on caller-supplied history IDs without verifying that the history row belongs to that prompt (history_entry.prompt_id == prompt.id). This affects /api/v1/prompts/id/{prompt_id}/history/diff, /api/v1/prompts/id/{prompt_id}/update/version, and /api/v1/prompts/id/{prompt_id}/history/{history_id}. An authenticated user with access to any prompt they control, plus a victim prompt_history.id, can read or delete another user's private prompt history. This vulnerability is fixed in 0.9.6.
Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to 0.9.6, a path traversal vulnerability exists in open-webui's cache file serving endpoint that allows any authenticated user to read files from sibling directories outside the intended cache directory, by exploiting an incomplete startswith containment check that lacks a trailing path separator. The root cause is that serve_cache_file() in open_webui/main.py validates the resolved path with file_path.startswith(os.path.abspath(CACHE_DIR)) — without appending os.sep. This allows any path resolving to a sibling directory whose name begins with cache (e.g. cache_sibling, cache_backup, cached_models) to pass validation. This vulnerability is fixed in 0.9.6.
Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to 0.9.6, Open WebUI patched SVG XSS in user profile images and webhook profile images but forgot to apply the same fix to model profile images. The ModelMeta class has no validate_profile_image_url field validator, and the model image serving endpoint has no MIME allowlist or nosniff header. Any authenticated user with workspace.models permission (enabled by default) can store a data:image/svg+xml;base64,... payload in a model's profile image and achieve full account takeover of anyone who navigates to the image URL. This vulnerability is fixed in 0.9.6.
Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to 0.9.6, Open WebUI lets a user who can create, update, or import workspace models store arbitrary meta.knowledge entries on their model without checking whether they own or can read the referenced files. Open WebUI then treats meta.knowledge entries of type file as an authorization source in two places: the built-in view_file tool reads the file's extracted text, and has_access_to_file()'s model branch authorizes the file content and file delete endpoints. A malicious model owner can therefore attach another user's file ID to their model metadata and read or delete that private file. This vulnerability is fixed in 0.9.6.
Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to 0.9.6,Open WebUI renders Mermaid blocks from Markdown files in the file preview panel and inserts the generated SVG into the DOM using innerHTML. Because Mermaid is configured with securityLevel: 'loose', attacker-controlled Mermaid content can be rendered unsafely in this flow. A working payload was validated through the Markdown preview path, resulting in JavaScript execution in the victim’s browser under the application origin. This vulnerability is fixed in 0.9.6.