CVE Database

Webhooks created by a collaborator keep firing after their repo access is revoked → ongoing real-time exfiltration of private repo content (incomplete revocation cleanup in `DeleteCollaboration`)

Gitea LFS Deploy-Key Privilege Escalation

Missing Authorization and Authorization Bypass Through User-Controlled Key and Incorrect Permission Assignment for Critical Resource and Exposure of Sensitive Information to an Unauthorized Actor in code.gitea.io/gitea

Public-only API token restriction is not enforced on team API routes

Public-Only Personal access tokens scope bypass in Organization and Permission Endpoints

Release attachment extension allowlist bypass via web release edit form (variant of CVE-2025-68939)

OAuth token introspection returns metadata of tokens issued to other clients (RFC 7662 section 4 violation)

Local File Inclusion via file:// URI in Migration Restore

Cross-Repo Information Disclosure via Org-Level Actions Run/Job APIs

Cross-repository issue/comment attachment re-linking can expose private attachment content

Denial of Service (CPU & Memory Exhaustion) via O(N^2) String Concatenation in Debian Package Upload

Gitea SSH Key Parser Denial of Service

Email Management API Bypasses ManageCredentials Feature Restrictions

CVE-2026-55402 is an out of bounds read vulnerability in Secure Access servers prior to version 14.57. Attackers with an ‘in the middle’ position can send specially crafted data to a server causing a persistent denial of service.

RSS/Atom feed handlers bypass API-token scope & public-only confinement (incomplete fix of #37698)

Denial of Service via Unbounded io.ReadAll in NPM Package Tag Endpoint

The GET /api/v1/user/actions/runners/registration-token endpoint (and its owner- and repository-level equivalents) creates a new runner registration token if none exists, yet the API scope middleware classifies it as read-only because it is a GET request. A holder of a leaked read:user-scoped token can therefore mint a registration token and register a malicious Actions runner that executes workflow jobs with access to repository secrets and source code.

Saurus CMS Community Edition contains an unauthenticated open redirect vulnerability in the logout handling code in classes/port.inc.php, where the url parameter supplied via GET or POST is passed directly to the Location header without domain allowlist, scheme validation, or relative path enforcement. Attackers can craft a malicious logout URL containing an arbitrary external domain or javascript: URI scheme to redirect authenticated users to attacker-controlled phishing pages after session destruction, enabling credential theft and OAuth redirect abuse.

In Zimbra Collaboration (ZCS) before 10.1.17, weak cryptographic key generation vulnerability exists in the OnlyOffice integration. The zimbraDocumentEditingJwtSecret is generated using an insecure random number generator, resulting in insufficient entropy. An attacker who obtains a JWT signed with the generated secret may be able to recover the JWT signing secret through offline brute-force, potentially enabling JWT forgery.

In Zimbra Collaboration (ZCS) before 10.1.17, a stored cross-site scripting (XSS) vulnerability exists in the Zimbra Classic Web Client due to insufficient sanitization of specific attachment content during inline preview. An attacker can send a crafted email containing a malicious attachment that, when previewed by a user, executes arbitrary JavaScript within the victim's browser session. Successful exploitation may allow an attacker to perform unauthorized actions on behalf of the victim user, potentially leading to data exfiltration or unauthorized access to sensitive information.

vLLM is an inference and serving engine for large language models. From 0.19.0 until 0.26.0, the /v1/completions CompletionRequest.prompt field in vllm/entrypoints/openai/completion/protocol.py accepts an unbounded list[str] or list[list[int]], prompt_to_seq() in vllm/renderers/inputs/preprocess.py and OnlineRenderer.preprocess_completion() in vllm/renderers/online_renderer.py expand every element, and vllm/entrypoints/openai/completion/serving.py creates one engine generator and response slot per prompt, allowing an authenticated API client to exhaust CPU, memory, async scheduling capacity, engine request slots, and response buffering with one request. This issue is fixed in version 0.26.0.

CVE-2026-55401 is a null dereference vulnerability on the load-balancing sub-system of Secure Access servers prior to 14.57. Attackers can send an unauthenticated packet to a Secure Access server with load balancing enabled, which results in the internal load balancer crashing. After a successful attack, the Secure Access server is still able to accept connections and is still able to issue a failover to connected clients. ‍ https://www.first.org/cvss/calculator/4.0#CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:N/VI:N/VA:L/SC:N/SI:N/SA:L

CVE-2026-55400 is an integer underflow in Secure Access servers prior to version 14.57. Attackers with an authenticated session can send specially crafted traffic to a server in a non-default configuration and cause a persistent denial of service.

Cross-site Scripting in the Markdown renderer in maalfer Pentestify before 2.3.2 allows authenticated users to execute arbitrary JavaScript in the application origin via a Markdown link whose URL contains a double quote, which closes the anchor's href attribute because the renderer's sanitization step does not escape quotes

Perl versions from 5.9.4 before 5.41.9 produce incorrect regular expression match results when a stale failure flag ends the Aho-Corasick prescan early in S_find_byclass. The prescan walks the subject for positions where the full pattern could match, and the engine tries it from the leftmost one recorded. A failing transition sets the failed flag, and a later successful transition does not clear it, so the prescan reads the stale flag as a failure and stops before it can record a candidate that starts earlier. It takes a subject where one candidate is recorded and a later character then forces a fallback through a fail link that succeeds. Example: "ABCDE" =~ m/ABCF|BCDE|C/; # matches C at offset 2, not BCDE "ABCDE" =~ m/ABCF|BCDE|C(G)/; # no match, BCDE missed An alternation like this can miss input it should match, or match it on the wrong branch, so an access or filtering decision made from the result can be wrong.

vLLM is an inference and serving engine for large language models. Prior to 0.27.0, an integer overflow in blockIdx.x * 2 * d in activation_kernels.cu can cause act_and_mul_kernel to consume another batched user's input, allowing a request processed in the same inference batch to receive a partial or complete copy of another user's inference result. This issue is fixed in version 0.27.0.

vLLM is an inference and serving engine for large language models. From 0.20.2rc0 until 0.26.0, safe_load_prompt_embeds in vllm/renderers/embed_utils.py uses torch.sparse.check_sparse_tensor_invariants, whose process-global save, enable, and restore state can be raced by concurrent prompt_embeds parts submitted to POST /v1/chat/completions through AsyncMultiModalItemTracker.resolve_items, asyncio.gather, and the default executor, allowing an invalid sparse tensor to reach tensor.to_dense despite the CVE-2025-62164 guard when enable_prompt_embeds is enabled. This issue is fixed in version 0.26.0.

vLLM is an inference and serving engine for large language models. Prior to 0.26.0, the structured_outputs.regex parameter in vllm/v1/structured_output/backend_lm_format_enforcer.py is passed to lmformatenforcer.RegexParser without compile_regex_with_timeout or validation in validate_structured_output_request_lm_format_enforcer, allowing an unauthenticated /v1/completions request against the lm-format-enforcer backend to consume a CPU core and stall the structured-output engine path with a catastrophic regular expression. This issue is fixed in version 0.26.0.

vLLM is an inference and serving engine for large language models. Prior to 0.26.0, the validation_exception_handler in vllm/entrypoints/openai/server_utils.py converts FastAPI RequestValidationError objects with str(exc), and sanitize_message in vllm/entrypoints/utils.py does not remove traceback-style file paths, allowing unauthenticated malformed JSON requests to /v1/chat/completions, /v1/completions, /tokenize, and /detokenize to disclose the OS username, home and virtual-environment paths, Python version, internal package structure, line numbers, and endpoint handler names. This issue is fixed in version 0.26.0.

Netty is an asynchronous, event-driven network application framework. Prior to 4.1.136.Final and 4.2.16.Final, io.netty.handler.codec.dns.AbstractDnsRecord, io.netty.handler.codec.dns.DefaultDnsRecordDecoder.decodeRecord(), and io.netty.handler.codec.dns.DnsCodecUtil.decompressDomainName() failed to release retained or newly allocated ByteBuf objects when IDN.toASCII() or encodeDomainName() rejected a malformed domain name, allowing unauthenticated remote DNS packets to leak direct memory incrementally until denial of service. This issue is fixed in versions 4.1.136.Final and 4.2.16.Final.