Docker

phpSysInfo up to and including version 3.4.5 exposes a complete IP allowlist bypass in read_config.php that any unauthenticated remote attacker can exploit by sending a spoofed X-Forwarded-For or Client-IP HTTP header set to any address present in the PSI_ALLOWED configuration. The flaw (CWE-290) renders the application's only access-control mechanism entirely ineffective, allowing full retrieval of system telemetry including hostname, kernel version, CPU model, memory layout, filesystems, and network interface addresses. A working proof-of-concept is publicly available on Exploit-DB; no CISA KEV listing exists at time of analysis.

Blind SQL injection in WsgiDAV's MySQLBrowserProvider sample module (versions <= 4.3.4) allows unauthenticated network attackers to read arbitrary data from the backing MySQL database via URL path manipulation. The record key segment of the URL is concatenated directly into a WHERE clause without escaping, enabling a boolean status-code oracle that supports full character-by-character data extraction. A secondary typo in the numeric-type check (`INTT` instead of `INT`) ensures all primary key types take the vulnerable quoted branch, and a working PoC with complete extraction script has been publicly confirmed. No public exploit identified as confirmed actively exploited (CISA KEV), but the PoC lowers the exploitation bar significantly.

NF registration poisoning in free5GC's Network Repository Function (NRF) lets attackers with SBI network access register forged Network Function profiles containing arbitrary, attacker-controlled service endpoint IPs, because the RegisterNFInstance handler performs no field validation against 3GPP TS 29.510. Legitimate NFs (AMF, SMF, AUSF, UDM, PCF, NSSF) then discover the fake profiles via NFDiscover and route control-plane signaling to the attacker, enabling interception, OAuth2 credential harvesting, and denial of service across the entire 5G core service mesh. Publicly available exploit code exists and the vendor has released a patch; the flaw is CWE-20 with a vendor CVSS 4.0 base score of 9.3 (Critical).

Incomplete authorization fix in Vikunja v2.3.0 allows Write-level collaborators to detach a shared child project from its parent hierarchy by supplying parent_project_id: 0, bypassing the Admin permission gate introduced by the CVE-2026-35595 fix. The flaw exists in versions v2.3.0 through the latest unstable builds and is fixed in v2.4.0. A detailed, independently reproducible proof-of-concept with curl commands is publicly documented in the GHSA advisory, though no CISA KEV listing has been identified at time of analysis.

Unauthenticated remote code execution in SENAITE.CORE (senaite.core) 2.0.0 through 2.6.0 lets any network-reachable attacker run arbitrary Python in the Zope worker process by chaining a missing authorization check with an eval() sink in the JSON API. The anonymous-reachable /@@API/update route calls eval() on attacker-supplied RecordsField/RecordField values before any permission or write check fires, and a public working PoC (tested against the stock senaite/senaite:v2.6.0 Docker image) demonstrates credential-free command execution. Publicly available exploit code exists; the flaw is not listed in CISA KEV, so there is no confirmation of active exploitation.

Remote code execution in qwed 5.1.1 allows any authenticated tenant to execute arbitrary Python code inside the API server process by submitting crafted mathematical expressions to the POST /verify/math or POST /verify/batch endpoints. User accounts are freely self-registerable via the default-enabled /auth/signup endpoint - no invitation code or administrator approval required - making the effective exploitation barrier nearly equivalent to unauthenticated. A complete, functional proof-of-concept exploit script is publicly available in the GitHub advisory (GHSA-q27q-98j4-9pfv), and a vendor-released patch is available in version 5.1.2.

Arbitrary OS command execution in mcp-shell's default Docker image allows any MCP tool caller to bypass the secure-mode executable allowlist by supplying `/bin/bash -c <payload>` to the `shell_exec` tool, running commands as `mcpuser` (UID 1000) inside the container. The validator in `security.go` extracts only the first whitespace-delimited token as the executable, approves `/bin/bash` against the default allowlist, and passes the `-c` flag through the metacharacter check unchallenged because the default `blocked_patterns` array is empty. A fully working PoC is included in the GitHub Security Advisory GHSA-3x77-wg38-92r3; no public exploit identified at time of analysis beyond the researcher-provided PoC, and no CISA KEV listing exists. Vendor-released patch v0.6.0 is confirmed.

OS command injection in mcp-shell's 'secure mode' allows any MCP client to execute arbitrary shell commands by exploiting Git's runtime alias mechanism. The security validator in security.go blocks common shell metacharacters but omits '!', the prefix Git uses to invoke shell aliases via '-c alias.NAME=!CMD'. Because /usr/bin/git is in the default allowed_executables list and the CVSS PR:N vector confirms no authentication is required, all default Docker deployments (mcpuser, UID 1000, Git pre-installed) are exploitable by any entity with MCP connectivity. A fully functional PoC is published in the GHSA advisory; no CISA KEV listing is present at time of analysis.

Insufficient access control granularity in docker-socket-proxy allows adjacent-network attackers without credentials to read arbitrary files and exfiltrate entire container filesystems via four unguarded Docker API endpoints. When the CONTAINERS environment variable is set, the HAProxy-based proxy fails to block /containers/{id}/archive, /containers/{id}/export, /containers/{id}/logs, and /containers/{id}/top, treating namespace-level permission as blanket endpoint permission. No active exploitation or confirmed public exploit code has been identified at time of analysis, though a researcher gist is referenced alongside the upstream fix in PR #183.

Cross-user memory disclosure and tampering in Headroom's LLM proxy (versions before 0.36.1) lets a caller read or overwrite any user's stored LLM memory by simply naming another user's identifier in the client-supplied x-headroom-user-id request header. The proxy trusts this header as the authenticated identity across the chat-completion and websocket data-plane paths in headroom/proxy/handlers/openai.py, with nothing binding the value to the actual caller. The console script binds loopback by default, but the shipped docker-compose.yml publishes the proxy on 0.0.0.0 with no required proxy token, so a stock Docker deployment exposes these routes to the network unauthenticated. Reported by VulnCheck; no public exploit identified at time of analysis.

Server-side request forgery in Headroom LLM proxy (versions before 0.36.1) allows unauthenticated remote attackers to redirect proxy upstream requests to arbitrary internal, loopback, link-local, or RFC 1918 destinations - including cloud instance metadata endpoints such as 169.254.169.254 - by supplying a crafted x-headroom-base-url request header. The proxy returns upstream responses verbatim to the caller, disclosing internal service content; critically, it also forwards the original Authorization header to the attacker-designated host, enabling credential harvesting. Exploitation is trivially achievable against deployments using the vendor-shipped docker-compose.yml, which binds 0.0.0.0 and enforces no HEADROOM_PROXY_TOKEN by default - a configuration weakness the server itself warns about at startup. No public exploit or KEV listing is identified at time of analysis; a vendor patch is available in v0.36.1.

Missing authorization in the Mailu admin REST API allows unauthenticated remote attackers to strip IP-based access restrictions from any existing user token or alter its comment field, effectively neutralizing a key access-control layer on API tokens. All Mailu deployments running versions prior to 2024.06.52 with the admin REST API explicitly enabled are affected. No public exploit code or CISA KEV listing is identified at time of analysis, though the network-accessible, zero-authentication attack vector (PR:N, AC:L) makes this straightforward to leverage when the API is reachable.

Path traversal in Dockge's stack name handling allows authenticated users to exfiltrate secrets from .env and Compose files belonging to any directory the server process can reach, and to recursively delete those directories. The flaw exists because validate() - which enforces a strict name allowlist - is only called from the save() code path; the socket handlers that service read and delete operations pass the raw user-supplied name directly to Stack.getStack without sanitization. Because Dockge routinely runs as root with access to the Docker socket, the reachable set encompasses the full host filesystem, not just managed stacks. No public exploit has been identified at time of analysis and no CISA KEV listing exists, but the bypass is mechanically trivial for any authenticated user.

Remote denial-of-service in vouch-proxy v0.47.2 and earlier crashes the server process with a single unauthenticated HTTP request. The /validate endpoint's multipart cookie reassembly routine in pkg/cookie/cookie.go parses the total part count directly from the attacker-controlled Cookie header name - an N value of 10 billion passed as VouchCookie_1of10000000000 triggers a make([]string, N) call requesting approximately 160 GB of heap, immediately raising a Go runtime fatal out-of-memory error. No authentication, session token, or prior interaction is required; a fully working PoC with Docker reproduction artifacts is publicly disclosed in the GitHub Security Advisory, though no CISA KEV listing is confirmed at time of analysis.

Rate limit bypass in Lemmy prior to 0.19.19 and 1.0.0-beta.1 allows unauthenticated remote attackers to circumvent all IP-based throttling by injecting a spoofed address as the first value of the X-Forwarded-For header. Because the bundled docker/nginx.conf forwards XFF headers using $proxy_add_x_forwarded_for rather than $remote_addr, attackers control the address that actix-web's ConnectionInfo::realip_remote_addr reads, receiving a fresh rate-limit bucket per request. No public exploit or CISA KEV listing exists, but exploitation is trivially automated and directly enables account registration spam, credential brute-forcing, comment spam, content scraping, bulk uploads, and repeated settings imports.

Uprobe-based gadget attachment in Inspektor Gadget (versions >= 0.27.0, < 0.53.1) can be weaponized by any unprivileged container to cause approximately 53 seconds of CPU burn in the IG daemon by supplying a crafted `/etc/ld.so.cache` file - no container capabilities required. Because IG uses fanotify hooks to pause container startup until uprobe attachment completes, and Docker serializes container starts behind that pause, a single crafted container effectively blocks all other containers on the host from launching for the duration of processing. The attack is amplifiable by starting multiple crafted containers simultaneously, and no public exploit has been identified at time of analysis.

Environment variable exposure in Renovate (versions 42.68.1 through 42.96.3 and 43.0.0 through 43.4.4) leaks the full process environment - including privileged credentials - into every child process spawned during dependency update operations. The regression was introduced inadvertently on 2025-12-30 when Renovate migrated to the `execa` library, whose default behavior extends rather than replaces environment variables, silently bypassing Renovate's intended allowlist filtering. Both insider attackers (with repository write access) and outsider attackers (via supply-chain compromise of a processed dependency) can embed credential-harvesting code in scripts that Renovate executes, exfiltrating GitHub tokens, registry credentials, and cloud IAM keys accessible to the deployment. No public exploit has been identified at time of analysis, but the vendor explicitly recommends rotating all credentials that affected Renovate instances had access to.

Arbitrary root file read in Linuxfabrik monitoring plugins allows a local attacker controlling the nagios or icinga account to exfiltrate any root-readable file by exploiting the hidden but permanently live --test argument accepted by all sudoers-whitelisted check plugins. Invoking the deb-updates plugin via sudo with --test=/etc/shadow,,0 causes lftest.py to read the supplied path as root with no path confinement and echo its full contents to stdout; approximately 22 additional whitelisted plugins yield filtered leaks or file-existence oracles through the same mechanism. No active exploitation has been confirmed by CISA KEV, but a fully functional proof-of-concept command is publicly disclosed in the vendor's own GitHub security advisory GHSA-rh9c-rqvg-f7pr, and patches are available in linuxfabrik-lib v6.1.0 and monitoring-plugins v7.0.0.

Path traversal via unsanitized plugin identifier in MineAdmin's app-store service enables any authenticated user to read directory existence outside the plugin root, trigger Plugin::install() and Plugin::uninstall() on arbitrary filesystem paths, and potentially execute arbitrary Composer commands. Affected versions are all releases of composer/mineadmin/mineadmin prior to v3.2.0-alpha.2. Compounded by a missing PermissionMiddleware on the plugin store controller (GM-4340), any bearer-token holder - not just administrators - can reach the vulnerable endpoints. A proof-of-concept is publicly available and the vulnerability has been reproduced in a Docker environment.

Uncontrolled decompression in Malcolm's upload-processing pipeline allows any authenticated user with upload permissions to exhaust the shared Docker volume by submitting a single-stream compression bomb (.gz, .bz2, .xz, .lzma, .lz). The flaw in scripts/safe-extract.py is a logic gap: entry-count, nesting-depth, and total-uncompressed-byte guards exist for container formats (zip/tar/rar/7z via libarchive) but are entirely absent for single-stream formats, allowing unbounded expansion. Successful exploitation denies service to all platform users simultaneously by starving OpenSearch, Logstash, Arkime, and Zeek of shared disk. No public exploit or CISA KEV listing has been identified at time of analysis; however, the attack is conceptually trivial once the bypass is known.

Denial-of-service in vm2 (Node.js sandbox library) versions 3.11.0-3.11.5 allows sandbox code to bypass the `bufferAllocLimit` DoS mitigation via two unwrapped allocation paths - `Buffer.concat(list, totalLength)` and `Buffer.from({length: N})` - committing unbounded host external memory in a single synchronous C++ call that V8's timeout mechanism cannot interrupt. A publicly available PoC confirms that a 200-byte sandbox payload inflates host RSS by hundreds of megabytes per invocation, making a single crafted request sufficient to OOM-kill a host process running inside a Docker container, Kubernetes pod, or AWS Lambda function. No public exploit identified in CISA KEV at time of analysis, though functional exploit code is embedded in the GHSA advisory itself.

Remote code execution in UpTrain 0.7.1 and prior allows authenticated users to execute arbitrary commands on the host system via the `/new_run` API endpoint. The `checks` and `metadata` parameters are unsanitized, enabling injection attacks (CWE-74) that run in the context of the underlying process - typically a Docker container per the official deployment documentation. Publicly available exploit code exists (CVSS 4.0 E:P per GitHub Security Lab advisory GHSL-2024-200/201), and no vendor-released patch has been identified at time of publication.

Remote code execution in UpTrain's `/add_prompts` API endpoint allows any authenticated user to execute arbitrary commands on the host system by injecting malicious payloads into the `checks` or `metadata` parameters. Affected versions are 0.7.1 and prior, and deployments following the official documentation run inside Docker containers, meaning successful exploitation grants control of the container environment. GitHub Security Lab (GHSL-2024-200/201) has published an advisory and the CVSS 4.0 E:P modifier confirms publicly available proof-of-concept exploit code exists; no vendor-released patch has been issued as of time of publication.

Remote code execution in UpTrain v0.7.1 and prior is achievable by any authenticated user through the `/create_project` API endpoint, where the `checks` and `metadata` parameters are processed without adequate sanitization, allowing arbitrary code to run in the context of the host process. Discovered and reported by GitHub Security Lab (GHSL-2024-198/GHSL-2024-199), the flaw targets the dashboard backend (app.py:L691) and is most commonly exploited within Docker-based deployments as recommended by official documentation. No vendor patch exists as of publication, and a proof-of-concept is confirmed by the CVSS 4.0 E:P modifier, elevating real-world urgency despite the authentication prerequisite.

Predictable temporary file paths in Etherpad's import and export handlers expose multi-tenant deployments to a symlink-based file overwrite attack. All ep_etherpad-lite versions through v3.0.0 are affected; exploitation requires local host access, an unprivileged OS account, and - for high-impact scenarios - the Etherpad process running as root, a common default in many Docker base images. A proof-of-concept sketch is included in the vendor advisory GHSA-2jwf-f4xq-f24h; no public weaponized exploit code beyond that sketch exists and this vulnerability is not listed in the CISA KEV catalog.

Remote command execution in Flowise <= 3.1.2 lets an authenticated user break out of the vm2/@flowiseai/nodevm JavaScript sandbox by abusing the /api/v1/node-custom-function endpoint. By passing attacker-controlled executablePath and args into puppeteer.launch(), the code reaches child_process.spawn() outside the sandbox, yielding arbitrary OS command execution as the Flowise process user (root in the official Docker image) and host file disclosure via Chromium's file:// handling. No public exploit is identified at time of analysis, but a vendor security advisory (GHSA-9gvv-qjj3-2p6g) and VulnCheck confirm the flaw; the CVSS 4.0 base score is 9.4.

Remote unauthenticated authentication bypass in WolfStack before 25.9.2 lets attackers fully compromise container hosts by supplying a hard-coded cluster secret in the X-WolfStack-Secret header. Because the secret is a compile-time constant published in src/auth/mod.rs, anyone who can reach a node's management port passes the require_auth() gate with no session, API key, or account, and can then enumerate every Docker/LXC container and run arbitrary commands as root inside any of them. No public exploit is identified at time of analysis, but the flaw is trivially reproducible from source and was reported by VulnCheck.

Command injection in stata-mcp (pip:stata-mcp < 1.19.0) enables full OS-level remote code execution against any principal that can invoke the `ado_package_install` MCP tool. The `package` argument is concatenated unsanitized into a Stata command string delivered via pexpect.sendline(), allowing embedded newline characters to inject arbitrary Stata commands - including the `shell` escape - that execute under the account running the MCP server. Because the tool is registered in the default `all` profile and `all` is the active default, no misconfiguration by the victim is required; a detailed proof-of-concept confirmed via Docker reproduction is publicly available in the GitHub advisory, and a vendor-released patch exists in v1.19.0.

URLSecurityValidator SSRF bypass in compliance-trestle <= 4.0.3 allows a network-positioned attacker to reach cloud metadata services, loopback admin interfaces, and RFC 1918 internal networks by embedding IPv4-mapped IPv6 literals (e.g., [::ffff:169.254.169.254]) or the unspecified address 0.0.0.0 in OSCAL artifact import references processed by HTTPSFetcher or SFTPFetcher. Python's ipaddress module silently returns False - rather than raising TypeError - when comparing IPv6Address objects against IPv4Network ranges, so the validator's membership checks pass these forms through to requests.get unchanged. A publicly available proof-of-concept reproduces end-to-end IMDS credential exfiltration; no active exploitation is listed in CISA KEV and EPSS stands at 0.42%.

Path traversal in JFrog Artifactory (CWE-22) enables an authenticated low-privilege user to write data outside the intended Docker cache directory when specific remote-repository conditions are present. The CVSS 3.1 vector scores 5.3 Medium, reflecting a real but constrained integrity risk - arbitrary filesystem writes are possible, but high attack complexity and required authentication limit opportunistic abuse. No active exploitation or public exploit code has been identified at time of analysis.

Docker Sandboxes (sbx) fails to enforce read-only semantics on virtio-fs host mounts: the in-guest bind correctly receives the read-only flag, but the underlying host-edge grant omitted from the policy-share allowlist carries no access mode, leaving the shared-export path fully writable at the host level. Low-privileged code executing inside the sandbox can enumerate or derive the virtio-fs export path and write to host directories the operator explicitly attached as read-only, breaking the isolation guarantee that is the product's core security promise. No public exploit has been identified at time of analysis, and the vulnerability is not listed in the CISA KEV catalog.

Sandbox escape in Cursor IDE for macOS (versions prior to 3.0.0) allows a malicious or prompt-injected AI agent operating in Auto-Run Sandbox mode to break container isolation by invoking the Dev Containers CLI to launch a privileged Docker container that mounts the host's virtiofs0 virtual filesystem. This grants the agent unrestricted read and write access to the user's entire home directory and enables arbitrary command execution at the user's privilege level - all without any additional permission prompt to the user. No public exploit code has been identified at time of analysis; a vendor-released patch is available as version 3.0.0.

Path traversal in XAgent's ToolServerNode workspace file endpoint allows any network-accessible user to read arbitrary files on the host by exploiting missing path containment validation in the `/workspace/file` handler. Because XAgent permits self-registration without email verification, an attacker requires no prior credentials - trivial account creation is sufficient to reach the vulnerable endpoint. Sensitive targets include application secrets, database credentials, and system configuration files that may be reachable through Docker volume mounts, with no public exploit identified at time of analysis but the fix commit publicly available.

Dozzle's webhook SSRF guard can be bypassed by authenticated users in versions 10.5.2 through 10.6.14 by supplying webhook URLs that resolve to IPv6 transition addresses encoding blocked IPv4 targets such as 127.0.0.1 or 169.254.169.254. The `isBlockedIP` function in `internal/notification/dispatcher/webhook.go` validated literal IPv4 and native IPv6 addresses but did not unwrap IPv4 addresses embedded inside 6to4, NAT64, Teredo, or IPv4-compatible IPv6 forms, making the guard trivially circumvented by encoding a loopback or link-local address in any of those transition formats. The CVSS 4.0 score is 2.3; publicly available exploit code exists per SSVC classification, though no active exploitation has been confirmed and the vulnerability is not in CISA KEV.

Portainer CE through 2.44.0 exposes root-level Docker host access to any authenticated low-privileged user via a URL path normalization flaw in the Docker proxy endpoint. The authorization middleware evaluates a non-canonical request path (e.g., percent-encoded slashes, double slashes, or traversal sequences) differently from the proxy layer that ultimately forwards it to the Docker daemon - allowing crafted requests to bypass all access controls while executing with full Docker API privileges. No active exploitation has been confirmed (no CISA KEV listing) and no public exploit code has been identified at time of analysis; EPSS data was not provided in the source intelligence.

Command injection in Dokploy (self-hosted PaaS) before 0.29.13 lets an authenticated low-privilege member run arbitrary OS commands on the control-plane host. The volumeName field in the volumeBackup.create and volumeBackup.runManually tRPC endpoints is interpolated unquoted into a shell string passed to child_process.exec; because the process holds Docker socket access, injected commands run with host/root-equivalent privileges. No public exploit is identified at time of analysis, but the fix is a one-character-class change and the vulnerable code path is well documented in the published PR and advisory.

Command injection in Dokploy's Docker build pipeline allows authenticated application editors to execute arbitrary OS commands on the Dokploy host by supplying shell metacharacters in the dockerContextPath build configuration field. The malicious path value flows from the dashboard UI through getDockerContextPath() directly into an unquoted cd shell command executed via execAsync, enabling full host-level code execution under the Dokploy server process. All Dokploy installations prior to version 0.29.13 are affected; no active exploitation or public exploit code has been identified at time of analysis, though exploitation requires only a low-privilege authenticated session.

Shell command injection in Dokploy exposes the underlying host OS to arbitrary command execution by any authenticated user with Compose service edit permissions. The flaw is in the sanitizeCommand function within packages/server/src/utils/builders/compose.ts, which stripped only whitespace and surrounding quotes before interpolating user-supplied compose.command values into live shell invocations such as docker ${command} - leaving semicolons, pipe operators, backticks, and subshell substitutions entirely intact. Working exploit payloads (semicolons, $(subshell), pipes, backtick substitution) are publicly available in the fix PR test suite at https://github.com/Dokploy/dokploy/pull/4863; no CISA KEV listing exists at time of analysis, and the vendor-confirmed patch is available in v0.29.13.

Incomplete authorization in Dokploy's WebSocket handlers prior to v0.29.13 allows any authenticated organization member to obtain root terminal access to servers and read logs or statistics for services they were explicitly denied access to. The four affected handlers (terminal.ts, docker-container-terminal.ts, docker-container-logs.ts, docker-stats.ts) verify organization membership but never call checkServiceAccess, getAccessibleServerIds, or accessedServices, nullifying all fine-grained role-based access controls. No public exploit or CISA KEV listing exists; however, the attack requires only a valid low-privilege organization account, making this a meaningful privilege escalation risk in any multi-tenant Dokploy deployment.

OS command injection in Dokploy self-hosted PaaS before 0.29.8 allows an authenticated user with project access to run arbitrary commands on the Dokploy host. The getRegistryCommands() function in packages/server/src/utils/cluster/upload.ts builds a `docker login` shell string by interpolating the attacker-controlled registry password and URL without escaping, so malicious credentials are executed when a swarm deployment runs. No public exploit identified at time of analysis, but the root cause and fix are fully disclosed in the vendor advisory (GHSA-prwq-2mcm-mvhr), PR #4579, and the v0.29.8 release.

OS command injection in Dokploy self-hostable PaaS versions prior to 0.29.13 lets an authenticated user with application create or update permission achieve arbitrary command execution on the build host. The dockerImage field is interpolated unquoted into shell commands in buildRemoteDocker() (packages/server/src/utils/providers/docker.ts) and validated only as an optional string, so shell command substitution such as $(...) or backticks runs on the local build host or a remote SSH build target. The upstream fix and its regression test publicly demonstrate working injection payloads, so publicly available exploit code exists, though there is no public exploit identified as weaponized and no CISA KEV listing; EPSS was not provided.

Cross-tenant remote command execution in Dokploy (self-hosted PaaS) before 0.29.13 lets any authenticated user holding only the server:read permission run arbitrary OS commands on servers belonging to other organizations. The swarm tRPC endpoints never verified that the supplied serverId belonged to the caller's active organization, and getNodeInfo interpolated an attacker-controlled nodeId straight into a remote shell command, chaining a broken multi-tenant authorization flaw with OS command injection. No public exploit or CISA KEV listing has been identified, though the vendor security advisory and a fix-verification test are public.

OS command injection in Dokploy prior to version 0.29.13 allows any authenticated user with project write access to execute arbitrary commands as the Dokploy server process. The buildRemoteDocker() function in packages/server/src/utils/providers/docker.ts interpolates the user-controlled dockerImage field directly into a shell command passed to execAsync() without sanitization. The same unsanitized interpolation pattern was present across three separate build pipeline files, indicating a systemic rather than isolated flaw. No public exploit has been identified at time of analysis, and the vulnerability is not listed in CISA KEV.

OS command injection in Dokploy (self-hosted PaaS) before 0.29.13 lets an authenticated user with the backup:restore permission run arbitrary commands on the Docker-privileged host. The backup.restoreBackupWithLogs tRPC subscription forwards an attacker-controlled databaseName into shell strings that Node.js exec passes to /bin/sh before docker exec, so the outer host shell expands injected metacharacters. Rated CVSS 9.9 with a scope change to the host; no public exploit identified at time of analysis, though the upstream fix PR ships regression tests containing working injection payloads.

Command injection in Dokploy self-hosted PaaS before 0.29.13 lets a low-privileged organization member run arbitrary OS commands inside the root Dokploy container, which holds the host Docker socket, effectively yielding host takeover. The destination.testConnection mutation interpolates unsanitized S3 fields (accessKey, secretAccessKey, region, endpoint, provider, bucket) into an rclone shell command run via child_process.exec. No public exploit is identified at time of analysis, but the fix is confirmed in v0.29.13 and the root cause is clearly documented in the upstream patch (PR #4873).

OS command injection in Dokploy self-hosted PaaS prior to 0.29.13 lets an authenticated project member with compose write and deploy permissions execute arbitrary commands on the Docker-privileged host. The compose.update operation persists an unvalidated composePath that is later interpolated verbatim into `docker compose -f`, `docker stack deploy -c`, and `touch` commands run via `/bin/sh -c`, so a crafted path fires shell metacharacters when the compose is deployed. No public exploit identified at time of analysis and the issue is not listed in CISA KEV, but the upstream regression test suite demonstrates working injection payloads, making the path to exploitation well-understood.

Privilege escalation via missing authorization in Dokploy, a self-hostable Platform-as-a-Service, lets any authenticated organization member open a root shell inside arbitrary containers on a self-hosted instance. The /docker-container-terminal WebSocket handler authenticated the caller but never checked the attacker-supplied containerId against the caller's role, organization, or service scope before invoking docker exec. Fixed in 0.29.13; no public exploit identified at time of analysis, though the upstream fix commit is public.

Privilege escalation to host root in Dokploy self-hosted PaaS versions prior to 0.29.13 allows any authenticated organization member to hijack WebSocket terminal and log-streaming handlers that authenticate the session but never check the role/permission model enforced elsewhere. Because these handlers can open an interactive shell into any container - including the Dokploy container that mounts the Docker socket - a low-privileged tenant can obtain root on the host and cross every tenant boundary. No public exploit identified at time of analysis, though the upstream fix commit and a regression test are public. The CVSS is 9.9 driven by scope change and full CIA impact.

Command injection in Dokploy self-hosted PaaS versions prior to 0.29.13 lets an authenticated user with database-service provisioning rights run arbitrary shell commands on managed remote servers. The user-controlled dockerImage field for PostgreSQL, MySQL, MariaDB, MongoDB, Redis, and libSQL services is interpolated unquoted into a 'docker pull ${dockerImage}' command executed over the remote-server code path (execAsyncRemote), enabling shell metacharacter injection. Rated CVSS 9.9 with a scope change; no public exploit or CISA KEV listing is identified, though the patch ships a regression test demonstrating the injection.

Command injection in Dokploy's `createCommand()` function allows authenticated users with compose deployment access to execute arbitrary OS commands on the underlying Docker host. All Dokploy releases before 0.29.13 are affected via CPE cpe:2.3:a:dokploy:dokploy:*. The official CVSS score of 6.5 with C:H/I:N/A:N materially understates actual impact - successful exploitation yields full host-level command execution, not merely information disclosure. No public exploit identified at time of analysis, though working injection payloads are published directly in the fix commit's test suite.

Remote command injection in Dokploy self-hosted PaaS before 0.29.13 allows an authenticated dashboard user to execute arbitrary shell commands on managed remote nodes. The registry credential-testing and Docker Swarm cluster-management endpoints interpolate user-controlled values (nodeId, imagePrefix/username driving the registry tag) into unquoted shell strings that run over SSH via execAsyncRemote. Rated CVSS 9.9 with scope change; no public standalone exploit is identified, though the public fix commit df2779e and advisory GHSA-4mfc-grxw-6858 disclose the exact vulnerable code paths and injection payloads.

REST API connection exhaustion in Klever-Go versions 1.7.14 through 1.7.17 allows unauthenticated remote attackers to render node API services unavailable by holding slow-header connections open until server file descriptors are exhausted. The root cause is Gin's Engine.Run convenience method, which delegates to Go's default http.ListenAndServe without any ReadHeaderTimeout, ReadTimeout, or MaxHeaderBytes configured, leaving incoming connections open indefinitely if the client never completes request headers. A proof-of-concept documented in GHSA-w4c6-7r69-w7j9 confirmed that 120 such connections caused all 20 of 20 legitimate API probes to fail with 'accept: too many open files' on v1.7.17; no CISA KEV listing was found, and no public exploit beyond the researcher's PoC is identified at time of analysis.

Unauthenticated arbitrary file write in Dinky v1.2.5 and its byte-identical development branch lets anyone who can reach the HTTP port (8888 by default) overwrite files as the Dinky service account, because the POST /download/uploadFromRsByLocal handler feeds a caller-supplied path straight into new File(path) and transferTo(dest) with no validation. The route is exempt from Sa-Token auth and guarded only by a header check against a dinkyToken whose default value is hardcoded in source and shipped to every install, so the guard is effectively no guard. The researcher (VulnCheck) demonstrated full remote code execution by shadowing Dinky.class on the writable, chmod-777 classpath and by planting attacker JavaScript in the served index.html; there is no public exploit identified at time of analysis and it is not in CISA KEV, but the disclosure describes a working demonstration.

Credential exfiltration via the tenant API in OpenReception appointment-booking-software prior to v1.0.2 allows any authenticated TENANT_ADMIN to retrieve the full PostgreSQL connection string - including plaintext password - for the shared database instance. In default docker-compose.prod.yml deployments, those credentials belong to the postgres superuser (rolsuper=true), granting the attacker read-write access to every tenant's database, all GLOBAL_ADMIN password hashes and session records, and superuser-level PostgreSQL functions (pg_read_server_files, COPY FROM PROGRAM) usable for container-level escalation. No public exploit code or CISA KEV entry exists at time of analysis; EPSS data was not provided, but the CVSS S:C scope change and the cross-tenant blast radius make this a high-priority patch target for any operator running the affected version.

Traefik's BasicAuth middleware in versions 3.6.11-3.6.24 and 3.7.0-3.7.9 permits an authenticated low-privilege user to forge arbitrary backend identities via a singleflight key collision race condition. By concatenating their known valid password with a retrieved stored bcrypt hash, an attacker crafts a singleflight deduplication key identical to a legitimate in-flight request's key; the middleware returns the legitimate request's successful authentication result to the attacker's request, then propagates the attacker's chosen unconfigured username to the backend as a trusted identity. A fully working proof-of-concept achieving 25-of-25 successful identity forgeries under default bcrypt cost is publicly available in the GitHub security advisory; no CISA KEV listing has been identified at time of analysis.

Cross-user HTTP response poisoning in Traefik's default reverse proxy allows an unauthenticated remote attacker using HTTP/2 or HTTP/3 to smuggle responses into other clients' backend connections, potentially delivering authenticated or private content from one user to another. All supported Traefik branches through v2.11.52, v3.6.23, and v3.7.8 are affected in default configuration when fronting Go net/http or gunicorn/Flask backends, which are confirmed 'armed' (respond to CONNECT with keep-alive non-2xx without draining the body). A proof-of-concept is publicly available demonstrating reliable cross-user poisoning including a cascade scenario where 8 of 8 sequential victims read responses that were not their own; no CISA KEV listing was present at time of analysis.

Unauthenticated denial-of-service in sgoudelis/Ground Station prior to 0.6.0 allows any network peer to forcibly terminate the application process by emitting a single Socket.IO event requiring no credentials. Authentication enforcement is disabled and a wildcard CORS policy is in place, exposing the service_control handler - which triggers process shutdown via a restart_service command - to any host with TCP access to port 7000. In Docker deployments, the attack can be repeated in a loop to sustain a persistent outage of all satellite-tracking sessions, SDR recording pipelines, demodulators, decoders, and rotator controllers. No public exploit or CISA KEV listing is identified at time of analysis.

Unauthenticated remote code execution in OpenChamber 1.11.7 lets remote attackers run arbitrary OS commands by POSTing to the /api/fs/exec endpoint, which hands input directly to Node.js spawn() with no allowlist, blocklist, or argument sanitization. Because the authentication middleware silently no-ops when UI_PASSWORD is unset - the state of the default Docker deployment - the endpoint is reachable without credentials and returns full stdout, stderr, and exit code to the attacker. Reported by VulnCheck; there is no public exploit identified at time of analysis and it is not listed in CISA KEV, though the flaw is trivially exploitable.

Remote command injection in livebook-dev Livebook's deployment setup command generator allows an authenticated high-privilege attacker to poison Docker and Fly.io setup commands displayed in the web interface, causing arbitrary shell code to execute on the machine of any user who copies and runs the generated command. Affected versions span 0.13.0 through 0.18.6 and 0.19.0 through 0.19.8. No public exploit has been identified at time of analysis, though the CVSS 4.0 score of 4.9 reflects the high privileges required and the indirect execution path through a secondary user's action.

Host filesystem read and write access in IBM Langflow OSS 1.0.0 through 1.10.3 is achievable by any authenticated low-privilege user via insufficient filtering of Docker volume-mount and device-mapping arguments passed to Docker-based MCP servers. By injecting unsanitized flags such as --volume into MCP server configuration, an attacker can bind-mount arbitrary host directories into a spawned container, exposing or tampering with sensitive files outside the intended Docker isolation boundary. No public exploit code has been identified at time of analysis and no CISA KEV listing exists, but the network-exploitable, low-complexity attack path and full CIA impact score (CVSS 8.8) make this a high-priority patch target for any organization running Langflow in Docker environments with externally accessible or multi-tenant access.

Stored cross-site scripting in Ghost CMS feature image captions enables any authenticated staff user to inject unsanitized HTML into the post revision history modal, hijacking the Ghost Admin session of another staff or administrator user who views that revision. Affects Ghost npm package versions v4.9.0 through v6.54.0. No public exploit identified at time of analysis and not listed in CISA KEV, but the privilege escalation potential - a lower-privileged staff member compromising a full administrator session - makes this higher-impact than the CVSS 4.3 score implies for multi-tenant or shared Ghost Admin environments.

Server-Side Request Forgery in Ghost CMS v6.26.0-v6.54.0 allows unauthenticated remote attackers to abuse features such as Webmentions to issue blind HTTP requests from the Ghost server into its internal network, bypassing perimeter controls without returning response content to the attacker. The flaw stems from insufficient URL validation in outbound HTTP request handling, enabling internal host probing across a changed security scope (CVSS S:C). No active exploitation (CISA KEV) or public POC has been identified at time of analysis; the vendor has released a fix in v6.54.1.

Missing authentication in the NASA-AMMOS ANMS reference implementation exposes the amp-manager REST API (CivetWeb, port 8089) directly on the host network, allowing any network-reachable client to command DTNMA agents without credentials. Because the shipped docker-compose.yml publishes the port and bypasses the CAM authentication gateway, unauthenticated remote attackers can enumerate agents, submit arbitrary EXECSET command sets to them, and clear stored reports. No public exploit identified at time of analysis; the flaw affects the reference/testbed implementation rather than onboard spacecraft flight software, though managed agents may represent real ground or space nodes depending on deployment.

Session fixation in Ghost Admin (v2.2.0-v6.54.0) allows an attacker who can plant a known session identifier in a victim's browser - via a secondary vulnerability on the same domain - to hijack an authenticated Ghost Admin session after the victim logs in. The root cause is that Ghost Admin's login flow did not call session regeneration, leaving the pre-login session ID valid post-authentication. Exploitation is chained and non-trivial, but the upstream fix (v6.54.1) is confirmed and straightforward to apply. No public exploit code and no CISA KEV listing identified at time of analysis.

Path traversal in Ghost CMS theme upload (CWE-22) allows an authenticated staff-level user to write arbitrary files outside the designated uploads directory, potentially altering the behavior of the entire Ghost installation. Affected versions span v0.10.0 through v6.54.0 of the npm/ghost package - a wide range covering essentially all modern deployments. No public exploit code or CISA KEV listing exists at time of analysis; however, the CVSS Scope:Changed metric reflects that a successful write can affect components beyond the theme storage boundary, elevating the real-world impact above the moderate base score suggests.

Path traversal in Ghost's database backup export endpoint allows Administrator-level users to remotely overwrite arbitrary files on the server filesystem, affecting all Ghost installations from version 1.20.1 through v6.54.0 (npm package pkg:npm/ghost). An authenticated admin can supply a crafted export filename containing path separators (e.g., '../../etc/crontab') when triggering a database export, causing Ghost to write the JSON export blob outside the intended directory to an attacker-controlled path. No public exploit code exists and this CVE does not appear in CISA KEV at the time of analysis; however, the CVSS Scope:Changed rating reflects that successful exploitation crosses the application's filesystem boundary, with integrity and availability consequences beyond Ghost's own data directory.

Blind SSRF in Ghost Admin's image-size fetching module allows any authenticated staff-level user to issue arbitrary HTTP GET requests from the Ghost server to internal network hosts. Affecting all Ghost npm releases from v0.10.0 through v6.54.0, the flaw resided in the image dimension detection code path, which used an unprotected generic HTTP client with no SSRF controls when fetching external images for certain formats requiring a full response buffer. No public exploit or CISA KEV listing exists; no EPSS data was provided. The vulnerability was fixed in v6.54.1 by routing fetches through Ghost's SSRF-protected `externalRequest` got-based client.

Ghost CMS subscription checkout accepts redemption of archived offers due to missing offer-status validation in the member router controller, affecting all self-hosted Ghost instances from v4.22.0 through v6.54.0. Unauthenticated users who possess or can discover an archived offer ID can submit a valid checkout request and obtain unauthorized subscription discounts or access at a revoked price point. No public exploit code has been identified and this CVE is absent from CISA KEV, but the direct financial impact on subscription-based Ghost sites makes it a meaningful business-logic risk for affected operators.

Unauthenticated OAuth2 access-token theft in Flowise (npm/flowise <= 3.1.2) stems from the token-refresh endpoint POST /api/v1/oauth2-credential/refresh/:credentialId being listed in WHITELIST_URLS, so it bypasses authentication entirely. Any remote party who knows or guesses a credential ID can force the server to decrypt the stored OAuth2 credential, refresh it against the provider using the victim's client secret, and receive the fresh access_token directly in the response body. No public exploit code has been identified, but the researcher (Shinobi Security) validated the missing-auth behavior in Docker, and a vendor-released patch is available in 3.1.3.

Remote code execution in FlowiseAI Flowise 3.1.2 lets an authenticated user abuse the 'SQLite Record Manager' node's `additionalConfig` input to override the target database path and write an attacker-shaped SQLite file to any location on disk. Because the official Docker image runs as root, the file can be planted at /etc/chromium/exploit.conf, and a second Upsert operation that launches Puppeteer/Chromium sources that .conf file and executes an embedded reverse-shell payload, yielding full application compromise as root. No public exploit is identified at time of analysis beyond the detailed reproduction steps in the elttam advisory; the issue is patched in 3.1.3.

Remote code execution as root in Flowise (npm packages flowise and flowise-components, versions 3.0.5 through 3.1.2) lets any authenticated user with a valid API key escape the NodeVM sandbox used by the custom-function feature and run arbitrary OS commands. The flaw stems from executeJavaScriptCode() merging caller-supplied nodeVMOptions over its secure defaults with a JavaScript spread, so an attacker can re-enable blocked builtins (child_process, fs) via require.builtin:["*"]. A detailed working exploit is published in the vendor's GHSA advisory, so publicly available exploit code exists; there is no CISA KEV listing, meaning no confirmed active exploitation at time of analysis.

Authenticated remote code execution in FlowiseAI Flowise (versions ≤ 3.1.2, npm packages flowise and flowise-components) arises from a vm2 JavaScript sandbox escape reachable through the AgentAsTool, ChatflowTool, and ExecuteFlow nodes, which still run untrusted code in the deprecated in-process vm2 sandbox (useSandbox: false). An attacker who can edit a chatflow injects code via an unvalidated baseURL URL-hash fragment and abuses moment's CVE-2022-24785 path-traversal (whose fix is bypassable inside the sandbox) to break out of vm2 and run arbitrary commands as the Flowise service account, fully compromising the host. A complete working exploit is published in the elttam/GitHub advisory, but the flaw is not in CISA KEV.

Authenticated remote code execution in FlowiseAI Flowise 3.1.2 and earlier lets any logged-in user escape the vm2 sandbox and run arbitrary OS commands on the host. The MySQL/Postgres/SQLite Record Manager and MySQL/Agent Memory nodes expose an unfiltered additionalConfig input passed directly into the TypeORM DataSource constructor, whose entities/subscribers/migrations options can load attacker-supplied JavaScript files. A full public reproduction (including a root reverse-shell payload) was published in the elttam advisory, and the vendor fixed it in 3.1.3.

OS command injection in OpenWrt's luci-app-dockerman grants any authenticated user holding only the read ACL full root-level code execution on the affected router. The vulnerable run_ttyd handler in the ucode docker_rpc.uc backend constructs a shell command from attacker-controlled request fields (id, cmd, uid) and passes them unsanitized to system() within the rpcd root context, reachable via a simple HTTP POST to /ubus. Only LuCI master and openwrt-25.12 snapshots after the JS-to-ucode backend conversion are affected; upstream patch commits exist but no released patched snapshot version was confirmed as of the advisory, and no public exploit is identified at time of analysis.

Null byte injection in the free5GC Authentication Server Function (AUSF) v4.2.1 and earlier allows unauthenticated remote attackers to crash Go's URL parser and force HTTP 500 responses on the authentication endpoint, denying authentication service to all legitimate 5G subscribers sharing that AUSF instance. A publicly available Python PoC is included in the vendor's own security advisory, and fuzzing demonstrated a 4.1% trigger rate across nearly 100,000 requests - confirming reliable, scalable exploitation against the default deployment with no special configuration required. No public exploit or CISA KEV listing is confirmed at time of analysis, but the trivially automatable nature of the attack makes it a material risk for 5G core operators.

Sensitive-credential disclosure leading to full node compromise in Pterodactyl Wings before v1.12.3 lets a low-privileged, authenticated panel user (a server owner or a subuser holding startup.update) smuggle a {{config.*}} placeholder through a user-editable egg variable, causing Wings to resolve it against its entire daemon configuration and write secrets - including the node daemon token, token_id, and Docker registry credentials - into a file the user can read via the file manager or SFTP. Because the daemon token is both the Panel⇆Wings API bearer and the HMAC signing key for every JWT the node issues, an attacker can then forge tokens and act against every server on the node. No public exploit identified at time of analysis and it is not listed in CISA KEV, but the CVSS is 9.9 and the affected templating pattern is common across stock and community eggs.

Exposure of a default-credentialed PostgreSQL database in the VPS.org one-click Supabase template allows remote attackers to connect directly to the database on 0.0.0.0:5432 using the well-known default password 'postgres', gaining full read/write access. The flaw stems from insecure default initialization combined with Docker's iptables rules overriding host UFW firewall policy, so operators who believe UFW protects the port are silently exposed. No public exploit has been identified at time of analysis, and EPSS is low (0.14%, 4th percentile), but SSVC rates the technical impact as total and exploitation as automatable.

Remote code execution in Somta Juggle through version 1.6.0 lets unauthenticated network attackers run arbitrary OS commands by reaching the bundled H2 database web console, which ships exposed at /h2-console with default credentials. Because the stock Docker image runs as root, successful exploitation via the classic H2 CREATE ALIAS Runtime.exec() technique yields root-level code execution on the host container. No public exploit identified at time of analysis, though the VulnCheck advisory provides a detailed exploitation path.

Predictable shared-secret generation in Spring Tools for Eclipse (5.2.0 and earlier) weakens the authentication protecting Spring Boot DevTools remote-restart uploads to remotely deployed targets such as Docker containers or Cloud Foundry apps. Because the secret is derived from a non-cryptographic PRNG rather than a secure random source, an attacker able to reach the deployed application's DevTools remote endpoint may predict the secret, authenticate restart uploads, and push arbitrary code, yielding remote code execution (CVSS 8.3, scope-changed). There is no public exploit identified at time of analysis, and it is not listed in CISA KEV.

Network exposure of Docker container control ports in Spring Tools for Eclipse (versions 5.2.0 and earlier) occurs because the Boot Dashboard Docker integration binds these ports to 0.0.0.0 (all interfaces) instead of loopback. Any attacker on the same network segment as a developer running the integration can reach the container control interface without authentication, yielding high confidentiality, integrity, and availability impact. There is no public exploit identified at time of analysis and the issue is not in CISA KEV, but the flaw is trivially triggered once a developer starts the affected feature.

Arbitrary host file write in termux/proot-distro before 5.1.5 lets a malicious rootfs tarball or Docker layer plant an absolute-path symlink during `proot-distro install`, so a follow-up file member is written through that symlink onto the host filesystem - outside the container, before it is ever run. The install/reset paths (_extract_plain_tar) and the Docker layer path (_apply_layer) validated member names for '..' traversal but never validated symlink targets in member.linkname. No public exploit identified at time of analysis and it is not in CISA KEV, but the flaw is confirmed and fully described by the vendor GHSA advisory with vulnerable source shown.

Authentication bypass in Cosmos-Server 0.22.18 and earlier allows any holder of a valid Constellation device API key to impersonate any user (including admin) to backend applications by sending crafted x-cosmos-user headers. The tokenMiddleware prematurely returns before stripping these headers when the request originates from the Constellation tunnel, bypassing JWT authentication, MFA, and the AdminOnly access gate. Publicly available exploit code exists but no active exploitation has been reported.

Authentication bypass in pytonapi 2.2.0 allows unauthenticated remote attackers to trigger webhook handlers when a custom path is configured via the documented `path=` argument. The TonapiWebhookDispatcher fails to store bearer tokens for custom paths, causing the token check to be skipped entirely. Publicly available exploit code exists but no active exploitation in the wild has been confirmed.

Remote unauthenticated denial of service in gopacket <= 1.6.0 via a single crafted sFlow UDP datagram that triggers an unbounded memory allocation of up to 16 GiB, causing the parsing process to be OOM-killed. Affected services include any network telemetry collector or application that uses gopacket’s sFlow decoder without additional validation. A proof-of-concept exploit exists and can crash the process with just a 104-byte packet.

Remote denial-of-service in the gopacket library's Diameter AVP decoder allows unauthenticated attackers to trigger a ~4 GiB memory allocation per crafted packet, causing OOM kill after two messages under memory constraints. The flaw affects gopacket ≤1.6.0 (latest before fix), and publicly available exploit code demonstrates end-to-end exploitation against a collector with a 256 MB memory cap. No active exploitation is currently reported.

OS command injection in localstack serverless-localstack up to version 1.4.0 allows a local low-privileged attacker to execute arbitrary operating system commands by manipulating the `custom.localstack.docker.compose_file` configuration argument processed by the Configuration Handler in `src/index.js`. A publicly available proof-of-concept exploit exists via GitHub issue #303, elevating practical risk beyond the moderate CVSS 4.0 base score of 4.8. No vendor patch has been released, as the project maintainer has not responded to the responsible disclosure reported through an issue report.

Server-side template injection in Tugtainer (self-hosted Docker container auto-update tool) versions before 1.30.2 lets any authenticated user achieve root-level OS command execution by injecting Jinja2 expressions into notification templates. The title_template and body_template fields are rendered by an unsandboxed jinja2.Environment, so template payloads escape into arbitrary code running as root inside the container. Rated CVSS 9.9 with a scope change; no public exploit identified at time of analysis, and it is not listed in CISA KEV.

Denial of service in PhpSpreadsheet (PHPOffice) versions 2.0.0 through 5.8.0 lets remote unauthenticated attackers crash a PHP worker with a ~1 KiB malformed XLS/OLE file. The library's OLERead::read() follows attacker-controlled OLE sector chains without cycle detection, so a sector that points back to itself makes the small-block depot loop append the same data until memory is exhausted. It is triggered during automatic format detection via Reader\Xls::canRead(), so any app that probes untrusted spreadsheet uploads is affected; a working proof of concept is published in the vendor advisory, though the flaw is not in CISA KEV.

Denial-of-service in PhpSpreadsheet's Gnumeric reader lets remote attackers crash PHP worker processes by uploading a tiny gzip-compressed .gnumeric file that decompresses to hundreds of megabytes. The reader's Gnumeric::canRead() calls gzdecode() with no decompressed-size cap during routine file-type detection, so a ~98 KB payload expanding to ~96 MiB exhausts a 64M memory limit and triggers a fatal error before validation. No public exploit is identified at time of analysis, but a working proof-of-concept is described in the advisory and the CVSS score is 7.5 (High) with A:H impact.

Server-side request forgery in Gitea before 1.27.0 lets attackers coerce the server into HTTP requests against internal hosts because its outbound IP allow-list classifier (IsGlobalUnicast && !IsPrivate) fails to block CGNAT (100.64.0.0/10) and nine IPv6 transition prefixes. An authenticated user can point a webhook or repo migration at an internal CGNAT service and read the full response (status, headers, and up to 1 MB body) rendered back on the hook detail page; separately, when OpenID sign-in is enabled the unauthenticated /user/login/openid discovery fetch uses http.DefaultClient with no IP filtering, giving anonymous internet callers a blind SSRF primitive. Publicly available exploit code exists (bundled run_poc.sh reproduces all three primitives), but no active exploitation is confirmed.