Deserialization

PHP object injection in the Elated-Themes Töbel WordPress theme (versions up to and including 1.8.1) allows remote attackers to trigger unsafe deserialization of attacker-controlled data, potentially leading to arbitrary code execution, file manipulation, or data tampering depending on available POP gadgets. Rated CVSS 8.1 (High) with no public exploit identified at time of analysis and no CISA KEV listing, though the network attack vector and lack of authentication requirement make it a meaningful risk to any WordPress site running the theme.

Object injection in the Elated-Themes Aperitif WordPress theme through version 1.6 allows remote attackers to trigger PHP deserialization of attacker-controlled data, potentially leading to code execution, file manipulation, or full site compromise when a suitable gadget chain is present. CVSS 8.1 reflects high impact across confidentiality, integrity, and availability, though attack complexity is rated High. No public exploit identified at time of analysis and the issue is not listed in CISA KEV.

Unsafe deserialization in FoundationAgents MetaGPT versions up to and including 0.8.2 allows a local low-privileged attacker to achieve confidentiality, integrity, and availability impact by manipulating the `mapping` argument passed to `Message.check_instruct_content` in `metagpt/schema.py`. Publicly available exploit code (POC) exists via a GitHub issue report, elevating practical risk despite the local-only attack vector. No vendor patch has been released - the project was notified via issue report but has not responded, leaving installations without a remediation path.

Remote code execution in Microsoft SharePoint Server (2016 Enterprise, 2019, and Subscription Edition) allows an authenticated attacker to execute arbitrary code on the server by submitting crafted serialized data that triggers unsafe deserialization. The CVSS 8.0 vector requires low privileges and user interaction, and no public exploit is identified at time of analysis. The flaw is significant because SharePoint servers typically run with high privileges inside enterprise environments and frequently host sensitive collaboration data.

Remote code execution in IBM WebSphere Application Server 9.0 and 8.5 allows authenticated attackers to abuse unsafe Java deserialization in the SAML Web Single Sign-On component to run arbitrary code via a crafted HTTP request combined with a gadget chain. The flaw carries a CVSS 8.5 with scope change, and while no public exploit has been identified at time of analysis, deserialization gadget chains for WebSphere are historically well-researched. IBM has released a patch via support advisory node/7274733.

Remote code execution in IBM WebSphere Application Server 9.0 and 8.5 arises from unsafe deserialization of untrusted data processed by JAX-WS endpoints that use WS-Security. Unauthenticated remote attackers who can reach a SOAP/JAX-WS endpoint may craft malicious serialized payloads to execute arbitrary code in the WebSphere server context. No public exploit identified at time of analysis, but the high CVSS (9.0) and scope-changed impact mean any exposed JAX-WS service is a meaningful target.

Remote code execution in AMD's AI Tensor Engine for ROCm (AITER) through version 0.1.14 allows unauthenticated network attackers to run arbitrary code on every inference worker in a distributed cluster by sending a malicious pickle payload to the ZMQ SUB socket consumed by MessageQueue.recv() in shm_broadcast.py. The vulnerability stems from unauthenticated, unvalidated pickle deserialization with no HMAC or format checks; no public exploit identified at time of analysis, but VulnCheck has published an advisory and AMD has merged an upstream fix.

Deserialization restriction bypass in QOS.CH Sarl logback-core affects all versions through 1.5.33, allowing unauthenticated network attackers with the ability to influence serialized data to instantiate Java Proxy objects via SimpleSocketServer or SimpleSSLSocketServer. Despite the 'RCE' tag in source intelligence, the vendor explicitly states that no practical path to remote code execution or significant privilege escalation has been identified - this is a security boundary bypass of the HardenedObjectInputStream defense mechanism, not a full compromise vector. A proof-of-concept exists (CVSS E:P), though CVSS 4.0 scores the overall risk at 2.9 due to high attack complexity and prerequisite deployment conditions.

Unauthenticated remote code execution in Dassault Systèmes Teamwork Cloud (No Magic Release 2022x-2026x) and Magic Collaboration Studio (CATIA Magic Release 2022x-2026x) arises from unsafe deserialization of attacker-controlled data. The CVSS 9.8 vector indicates a network-reachable attack with no privileges or user interaction, yielding full confidentiality, integrity, and availability impact, though no public exploit identified at time of analysis and EPSS data was not provided.

Arbitrary code execution in ESA AnomalyMatch before 1.3.1 allows local attackers with low privileges to run code under the application's process by planting a malicious PyTorch checkpoint into a session directory, which is loaded via torch.load() with weights_only=False. No public exploit is identified at time of analysis, but the upstream fix (PR #9) and a third-party advisory (imlabs.info) confirm the unsafe-deserialization root cause and the migration to safetensors.

{/, l, o, g} - rather than a literal prefix strip, causing the filename derived from the URL to diverge from the file actually served. No public exploit has been identified at time of analysis, and this vulnerability is not listed in CISA KEV.

The event log detail endpoint in Apache Airflow before 3.2.2 applies a generic DAG-level audit log permission check rather than scoping authorization to the specific DAG that owns the requested event log entry, allowing any authenticated low-privilege user to read audit log entries belonging to DAGs outside their permitted scope. The flaw is a broken object-level authorization (IDOR) pattern - classified as CWE-639 - where the user-supplied `event_log_id` path parameter can reference log rows from unauthorized DAGs without triggering a rejection. No public exploit code exists and the issue is not listed in CISA KEV, but the attack is trivially executable by any authenticated Airflow user in a multi-tenant deployment.

Arbitrary Python module import in Apache Airflow versions prior to 3.2.2 occurs when the scheduler deserializes custom DeadlineReference objects, because the prior implementation called import_string() directly on an attacker-controllable __class_path field. Rated CVSS 7.3 with low confidentiality/integrity/availability impact, this issue has no public exploit identified at time of analysis and EPSS estimates exploitation probability at 0.02% (6th percentile).

Authenticated remote code execution in Apache Airflow 3.2.0 through 3.2.1 allows users with permission to update XCom entries to achieve code execution by submitting reserved deserialization metadata keys (e.g. __classname__, __type, __data__, __var) to the PATCH XCom endpoint. The XComUpdateBody datamodel omitted the FORBIDDEN_XCOM_KEYS validator that XComCreateBody enforced, letting attackers smuggle a malicious typed payload that is later deserialized into an arbitrary Python class. No public exploit identified at time of analysis and EPSS risk is negligible (0.02%), but a vendor fix has shipped and the root cause is a classic CWE-502 untrusted deserialization.

{method_name} and /simple_execute/{method_name} endpoints, which call pickle.loads() on raw HTTP request bodies. The flaw scored CVSS 4.0 of 9.2 and has an upstream fix in commit d7441481, but no public exploit was identified at time of analysis; risk is amplified by the default Docker image running as root, leading to full container compromise.

Authenticated PHP Object Injection in the WooCommerce Infinite Scroll and Ajax Pagination WordPress plugin (versions up to and including 1.8) allows Subscriber-level users to deserialize attacker-controlled data via the 'settings' parameter of the import_settings function. While the plugin itself contains no usable POP chain, the presence of any vulnerable gadget in another installed plugin or theme can escalate this into arbitrary file deletion, sensitive data disclosure, or remote code execution. There is no public exploit identified at time of analysis, but the low privilege barrier and ubiquity of WordPress gadget chains make this a meaningful risk for multi-plugin sites.

Security restriction bypass in logback-core's HardenedObjectInputStream allows limited object injection via logback's SimpleSocketServer and SimpleSSLSocketServer components, affecting all versions through 1.5.32 inclusive. An attacker who can influence serialized data submitted to these socket server endpoints can instantiate objects from java.lang and java.util classes not explicitly blocked by the hardened deserializer, circumventing its intended allowlist controls. The vendor and NVD both confirm no practical remote code execution or significant privilege escalation has been identified; the real-world impact is limited confidentiality and integrity exposure. No public exploit identified at time of analysis beyond E:P proof-of-concept maturity indicated in the CVSS vector. Not listed in CISA KEV.

Remote code execution in SMSGate sms-core versions 2.1.13.6 and earlier allows remote attackers to execute arbitrary code by sending crafted input to the Cmpp7FDeliverRequestMessageCodec.java component, which handles CMPP protocol message decoding. The CVSS 7.3 (AV:N/AC:L/PR:N/UI:N) vector indicates network-reachable, unauthenticated exploitation with low complexity, though EPSS scores this at only 0.06% (18th percentile) and there is no public exploit identified at time of analysis. SSVC indicates exploitation status is 'none' but the issue is automatable with partial technical impact across CIA.

Arbitrary file deletion in the WP Contact Form 7 DB Handler WordPress plugin (versions up to and including 3.0) can be achieved by chaining CSRF, UNION-based SQL injection, and PHP object deserialization. A remote unauthenticated attacker who lures a logged-in administrator to a malicious page can delete arbitrary server files, including wp-config.php, which typically forces the site into a re-installation state and enables full site takeover. No public exploit identified at time of analysis, though Wordfence's detailed write-up effectively documents the exploit chain.

Unauthenticated PHP object deserialization in Symfony's Monolog Bridge affects the development-time `server:log` console command, which by default binds its TCP log listener to 0.0.0.0:9911 and passes every received frame through `unserialize(base64_decode(...))` with no allowed_classes allowlist, authentication, or integrity check. Any host that can reach port 9911 on a machine running `server:log` can crash the listener (unauthenticated DoS) or trigger PHP object injection whose magic-method side effects may reach remote code execution when a usable gadget chain exists in the target's autoload set. No public exploit identified at time of analysis; EPSS is low (0.99%, 77th percentile) and the flaw is not in CISA KEV, consistent with a dev-only tool rather than a production-facing service.

Remote code execution in RELATE LMS (the inducer/relate web courseware platform) stems from its Celery task queue being configured to accept and unpickle untrusted messages (CELERY_ACCEPT_CONTENT included "pickle"). Because the code-execution sandbox lacks network isolation, an authenticated student can reach the message broker and deliver a malicious pickle payload that the worker deserializes, yielding arbitrary command execution on the host. No public exploit identified at time of analysis; the issue is corrected in commit d66ba5659b459bf1ba56b7109b5f9ecf197cbefb.

PHP object injection in Pimcore (packages pimcore/pimcore and admin-ui-classic-bundle) up to and including version 12.3.6 arises from six code paths calling unserialize() without the allowed_classes restriction on values read from database columns and filesystem files. An attacker who can already write to one of those sources - for example through SQL injection into the tmp_store, sites, or custom_layouts tables, or a file write to the WebDAV delete log - can plant a serialized PHP gadget chain that executes arbitrary code with web-server privileges once the data is deserialized. No public exploit identified at time of analysis (the vendor advisory documents only a conceptual PoC procedure), the CVE is not in CISA KEV, and EPSS is not provided; the issue is fixed in 12.3.7 and rated CVSS 8.0, with the High attack-complexity reflecting its dependence on a separate write primitive and a working gadget chain.

Unsafe deserialization in Jenkins Active Directory Plugin 2.41 and earlier allows a remote attacker holding administrative credentials to achieve full system compromise by manipulating the LDAP referral processing path. The plugin deserializes data received from LDAP referrals without validation (CWE-502), which can enable arbitrary code execution on the Jenkins controller. No public exploit exists at time of analysis, and CISA SSVC assesses this as not automatable, though technical impact is rated total - making it a targeted rather than opportunistic threat.

Jenkins LDAP Plugin versions up to and including 807.v7d7de30930cf deserializes Java objects returned via LDAP referral responses without any validation, exposing the underlying Jenkins instance to potential remote code execution via classic Java deserialization gadget chains. Exploitation is constrained by a high privilege requirement and high attack complexity (CVSS PR:H/AC:H), limiting realistic scenarios to attackers who already hold Jenkins administrative credentials or can manipulate LDAP referral destinations. No public exploit code has been identified and this vulnerability does not appear in the CISA KEV catalog at time of analysis.

SQL injection in Pimcore's admin-ui-classic-bundle (versions <= 2.3.5) allows an authenticated user holding only the translations-view permission to read arbitrary database contents by injecting into the translation grid's date filter. The user-controlled 'property' field of the filter JSON is interpolated directly into a UNIX_TIMESTAMP(DATE(FROM_UNIXTIME(...))) expression at the POST /admin/translation/translations endpoint, behind only a trivially bypassable str_replace('--','') filter. A working proof-of-concept and publicly available exploit code exist; the reporter notes it can be chained with an unsafe-unserialize flaw (GM-249) to reach remote code execution. No EPSS score or CISA KEV listing was supplied.

Insecure deserialization in NVIDIA Merlin Transformers4Rec on Linux allows a local attacker to achieve code execution, data tampering, and information disclosure by tricking a user into loading a malicious serialized object. The flaw affects all Main-branch commits prior to March 11, 2026, and currently has no public exploit identified at time of analysis, with a very low EPSS score (0.02%) reflecting limited real-world activity. CISA SSVC classifies exploitation as 'none' but technical impact as 'total', placing it firmly in the supply-chain/MLOps risk category rather than a mass-exploitation threat.

Remote code execution in Mirasvit Full Page Cache Warmer for Magento 2 before 1.11.12 allows unauthenticated attackers to execute arbitrary code by sending a crafted serialized PHP object in the CacheWarmer cookie. The flaw is confirmed actively exploited (CISA KEV) with publicly available exploit code, and successful exploitation chains Magento and dependency gadget chains via an unsafe call to unserialize(). Despite a low EPSS score (0.10%), KEV listing and CVSS 9.3 indicate this is a high-priority patch for any Magento 2 store running the module.

Unsafe deserialization in changmingxie tcc-transaction (versions up to 2.1.0) allows a remotely authenticated attacker with low privileges to exploit the Fastjson AutoType feature via the REST API, achieving limited confidentiality, integrity, and availability impact on the affected system. A proof-of-concept exploit exists (CVSS 4.0 E:P), referenced in a public GitHub bug report, though EPSS probability sits at just 0.04% (12th percentile) and SSVC assesses exploitation as none at time of analysis, indicating no observed active abuse. The vendor was notified prior to disclosure but did not respond, meaning no official patch has been released.

Remote code execution in HuggingFace Transformers prior to 5.3.0 allows attackers to achieve arbitrary code execution on a victim's machine by publishing a malicious model whose config.json sets the `_attn_implementation_internal` field to an attacker-controlled Hub repository. When the victim calls the standard `AutoModelForCausalLM.from_pretrained()` API, the library silently downloads and executes Python kernels from that repository with the victim's privileges, bypassing the `trust_remote_code` safety gate. No public exploit is identified at time of analysis (EPSS 0.03%, SSVC exploitation: none), but the technical impact is total and the attack uses the documented, default usage pattern.

Unsafe deserialization in Microsoft Planetary Computer Pro (Geocatalog) lets a remote unauthenticated attacker craft malicious serialized payloads that the service processes, resulting in information disclosure across a trust boundary. The maximum CVSS 10.0 score reflects network-reachable exploitation with no privileges or user interaction and a scope change, though no public exploit identified at time of analysis and EPSS data was not provided.

Authenticated remote code execution in Microsoft SharePoint (Enterprise Server 2016, Server 2019, and Subscription Edition) stems from unsafe deserialization of untrusted data (CWE-502), enabling an authorized attacker to run arbitrary code on the server over the network. CVSS 8.8 with low privileges required and no user interaction makes this attractive to post-authentication adversaries, though no public exploit identified at time of analysis and CVSS temporal data marks exploit code maturity as Unproven.

Arbitrary code execution in Amazon Braket Python SDK versions prior to 1.117.0 allows an authenticated attacker with S3 write access to the job output bucket to compromise any client machine that processes those job results. The flaw stems from insecure pickle deserialization in the job results processing component, and while no public exploit has been identified at time of analysis, the impact extends to every downstream consumer of poisoned results. EPSS data is unavailable, but the supply-chain-style propagation across analyst workstations and CI systems materially raises real-world risk.

Constraint extension stripping in the golang.org/x/crypto SSH agent client (versions prior to 0.52.0) allows remote SSH hosts to use forwarded keys without the destination restrictions the user intended. When clients added keys to a remote agent, extensions such as restrict-destination-v00@openssh.com were silently dropped during serialization, effectively converting scoped keys into unrestricted ones on downstream hosts. No public exploit identified at time of analysis and EPSS is very low (0.02%), but SSVC rates technical impact as total and automatable.

Remote code execution in Concrete CMS versions 5.0 through 9.5.0 allows a high-privileged administrator to bypass the platform's `_fromCIF` deserialization guard by submitting malicious payloads through the REST API instead of standard form POST requests. The flaw resides in the ExpressEntryList block controller (CWE-502) and stores a serialized PHP gadget in the `filterFields` database column, which is unmarshalled when another administrator subsequently views or edits the block, leading to full server takeover. No public exploit identified at time of analysis, and the issue is not present in CISA KEV.

Deserialization of untrusted data in Apache Fory PyFory. PyFory's ReduceSerializer could bypass documented DeserializationPolicy validation hooks during reduce-state restoration and global-name resolution. An application is vulnerable if it deserializes attacker-controlled data using PyFory Python-native mode with strict mode disabled and relies on DeserializationPolicy to restrict unsafe classes, functions, or module attributes. This issue affects Apache Fory: from before 1.0.0. Mitigation: Users of Apache Fory are recommended to upgrade to version 1.0.0 or later, which enforces DeserializationPolicy validation for the affected ReduceSerializer paths and thus fixes this issue.

Arbitrary code execution in NVIDIA BioNemo Framework on Linux allows a local attacker to abuse unsafe deserialization of untrusted data (CWE-502), leading to code execution, denial of service, information disclosure, and data tampering. The CVSS 7.8 vector indicates local attack vector with required user interaction, and no public exploit has been identified at time of analysis.

Unsafe deserialization in NVIDIA TensorRT-LLM's RPC testing component allows a local high-privileged attacker to trigger code execution, denial of service, data tampering, or information disclosure across a changed scope. The flaw is rated CVSS 7.5 despite local-only access and high attack complexity because successful exploitation crosses a security boundary (S:C) and yields full CIA impact. No public exploit identified at time of analysis, and the issue is not listed in CISA KEV.

Deserialization of untrusted data in NVIDIA TensorRT-LLM across all platforms allows a local, low-privileged attacker to achieve code execution, data tampering, and information disclosure by exploiting an unsafe serialized handle. The CVSS Changed Scope (S:C) indicates the impact can extend beyond the vulnerable component itself - notable given TensorRT-LLM's role as an inference serving library often integrated into multi-tenant or production AI infrastructure. No public exploit identified at time of analysis, and the vulnerability is not listed in the CISA KEV catalog.

Unsafe deserialization in NVIDIA TensorRT-LLM's MPI server component allows a high-privileged local attacker to achieve code execution, denial of service, data tampering, or information disclosure on systems running the affected library. The CVSS 7.5 score reflects high impact but constrained exploitability (AV:L/AC:H/PR:H), and no public exploit identified at time of analysis. Scope change (S:C) indicates compromise can extend beyond the vulnerable component to impact other resources on the host.

PHP Object Injection in the Boost plugin for WordPress (versions up to and including 2.0.3) allows unauthenticated remote attackers to inject arbitrary PHP objects via the STYXKEY-BOOST_USER_LOCATION cookie. The vulnerability stems from unsafe deserialization of attacker-controlled cookie data; while the plugin itself ships no usable POP (property-oriented programming) chain, exploitation becomes high-impact when any other installed plugin or theme provides one. No public exploit identified at time of analysis, and the issue is not currently listed in CISA KEV.

Java Deserialisation Vulnerability in Jaspersoft Reports Library leads to Remote Code Execution (RCE), potentially allowing code execution on the affected system

Unauthenticated root-level remote code execution affects HestiaCP versions 1.9.0 through 1.9.4 when the optional web terminal feature is enabled, stemming from a session-handling format mismatch (CWE-502) between the PHP backend and the Node.js web terminal. Remote attackers can inject crafted HTTP header data that PHP writes into session storage but Node.js parses with naive string splitting, yielding arbitrary command execution as root; no public exploit identified at time of analysis, though VulnCheck has published a technical advisory and the upstream patch is publicly diffable.

Remote code execution in the TYPO3 'Content Element Selector' extension allows unauthenticated attackers to execute arbitrary PHP code by sending a crafted cookie that the extension feeds directly into PHP's unserialize(). The flaw (CWE-502, CVSS 4.0 score 9.2) is exploitable only on installations where a content element is configured with 'Persistent Mode: Static'. No public exploit identified at time of analysis, though the deserialization pattern is well-understood and typically rapid to weaponize.

Remote code execution in the TYPO3 Crawler extension occurs when the X-T3Crawler-Meta response header from a crawled URL is passed unchecked to PHP's unserialize(), enabling arbitrary PHP object injection. Exploitation requires a high-privileged administrator to configure a crawler-enabled page and a Scheduler task pointing at an attacker-controlled endpoint, so while impact is full RCE on the TYPO3 host, it is gated by an unusual combination of admin access, user interaction, and externally reachable malicious URLs. No public exploit identified at time of analysis and no CISA KEV listing.

Insecure deserialization in Significant-Gravitas AutoGPT platform versions 0.6.34 through 0.6.51 lets an attacker who can poison entries in the shared Redis cache achieve arbitrary command execution inside the backend container. The backend's read path invokes pickle.loads on cache bytes with no HMAC, signature, or schema gate, so any attacker-controlled value reaching that key becomes code on retrieval. No public exploit identified at time of analysis, and the issue is not listed in CISA KEV; the vendor shipped a fix in autogpt-platform-beta-v0.6.52.

Remote code execution in APScheduler (all versions through 3.10.x and 4.0.0a5) is achievable when applications deserialize attacker-controlled data via the bundled JSONSerializer or CBORSerializer. The unmarshal_object routine dynamically imports modules and invokes __setstate__ on arbitrary classes, letting an attacker pivot an untrusted payload into code execution; publicly available exploit code exists, though EPSS remains low at 0.06% (19th percentile).

Remote code execution in FreePBX versions below 16.0.71 and 17.0.6 allows authenticated low-privileged users with backup access to execute arbitrary PHP code by uploading a malicious tar archive containing a crafted manifest file. The backup module passes attacker-controlled data directly to PHP's unserialize() without class restrictions, enabling PHP object injection that runs as the asterisk or www-data web server user. No public exploit identified at time of analysis, though the upstream fix commits are public on GitHub, making patch-diff exploitation feasible.

SGLangs multimodal generation runtime is vulnerable to unauthenticated remote code execution when the --enable-custom-logit-processor option is enabled, as Python objects loaded via dill.loads() will be deserialized without validation.

SGLangs multimodal generation runtime scheduler's ROUTER socket binds to 0.0.0.0 by default and contains a sink that calls pickle.loads() on incoming messages, enabling RCE when exposed to the internet.

Deserialization vulnerability in H2O-3 machine learning platform versions up to 7402 enables remote code execution through the importBinaryModel function when processing malicious JAR files. The vulnerability allows unauthenticated remote attackers to execute arbitrary code with publicly available exploit code (CVSS 7.3, EPSS not provided). The vendor failed to respond to disclosure attempts, leaving users without an official patch.

Unsafe deserialization in Oinone Pamirs versions up to 7.2.0 allows authenticated remote attackers to potentially execute arbitrary code via crafted JSON payloads to the appConfigQuery interface. The vulnerability exists in JsonUtils.parseMap within PamirsParserConfig.java, where attacker-controlled data is deserialized without proper validation. Public exploit code is available on GitHub, though EPSS and KEV data are not provided. CVSS 4.0 score of 2.1 reflects limited scope impact (VC:L/VI:L/VA:L with SC:N/SI:N/SA:N), requiring low-privilege authentication (PR:L) but featuring low attack complexity (AC:L) and network attack vector (AV:N). Vendor non-responsive to disclosure.

python jsonpickle 2.0.0 contains a remote code execution vulnerability that allows attackers to execute arbitrary Python commands by deserializing malicious JSON payloads containing py/repr objects. Rated critical severity (CVSS 9.3), this vulnerability is remotely exploitable, no authentication required, low attack complexity. Public exploit code available and no vendor patch available.

Path traversal in SimpleSAMLphp's CAS server module allows unauthenticated remote attackers to read and deserialize arbitrary files outside the ticket directory via crafted ticket parameters. When using FileSystemTicketStore, attackers can inject '../' sequences into CAS validation endpoints to escape the configured directory, potentially deleting files that contain serialized PHP data compatible with array types. The vulnerability has a CVSS score of 8.6 with no public exploits identified at time of analysis.

DataHub frontend versions prior to 1.5.0.3 deserialize untrusted Java objects from the REDIRECT_URL HTTP cookie during OIDC callback flow without integrity protection, allowing authenticated attackers to read sensitive information. The vulnerability affects the GET /callback/oidc endpoint and requires a valid OIDC identity provider account to exploit. A vendor-released patch is available in version 1.5.0.3.

Denial-of-service in GitLab Enterprise Edition allows a crafted file upload to exhaust service availability through improper deserialization validation. The vulnerability spans an exceptionally wide range, affecting all GitLab EE instances from version 11.9 through the 18.11 line until patched releases. There is no public exploit identified at time of analysis and EPSS sits at 0.02% (4th percentile), indicating low observed exploitation pressure, though the breadth of the affected version range means unpatched installations represent a meaningful attack surface for availability disruption.

Unsafe deserialization in LangSmith SDK's prompt pull methods allows remote attackers to execute server-side request forgery (SSRF) and redirect LLM traffic to attacker-controlled infrastructure when applications pull public prompts from LangSmith Hub. The SDK deserializes untrusted prompt manifests containing serialized LangChain objects with attacker-controlled constructor arguments, including malicious base_url configurations, custom headers, and secret references. Exploitation requires user interaction (developers must call pull_prompt with a malicious owner/name identifier), but no authentication is required to publish malicious prompts to the public Hub. Vendor-released patches in Python >= 0.8.0 and JS/TS >= 0.6.0 now block public prompt pulling by default, requiring explicit opt-in via dangerously_pull_public_prompt flag. EPSS data not available; no CISA KEV listing or public exploit identified at time of analysis.

Remote code execution in F5 BIG-IP and BIG-IQ Configuration utility allows authenticated attackers with low privileges to execute arbitrary code with high impact to confidentiality, integrity, and availability. The vulnerability stems from unsafe deserialization (CWE-502) in the management interface, exploitable over the network with low attack complexity and no user interaction required. Vendor-released patch available per F5 advisory K000156761. No public exploit identified at time of analysis, with CVSS 8.8 indicating critical severity for environments where attackers have valid low-privilege credentials to the Configuration utility.

PHP Object Injection vulnerability in coreActivity activity logging plugin through version 3.0 allows remote attackers to trigger persistent Denial of Service blocking administrator access to log pages. Unauthenticated attackers inject crafted PHP serialized payloads via User-Agent headers during any logged event (e.g., failed login). When administrators view the Logs page, the plugin deserializes untrusted data and passes it to DeviceDetector::setUserAgent(), causing Fatal TypeError. Vendor-released patch version 3.1 available (released May 6, 2026). EPSS exploitation probability not available; no CISA KEV listing at time of analysis. CVSS 8.1 reflects high complexity attack requiring precise payload crafting despite no authentication requirement.

Adobe Connect versions 2025.9.15, 2025.8.157 and earlier are affected by a Deserialization of Untrusted Data vulnerability that could result in arbitrary code execution in the context of the current user. An attacker could exploit this vulnerability to execute arbitrary code. Exploitation of this issue requires user interaction in that a victim must visit a maliciously crafted URL or interact with a compromised web page. Scope is changed.

Deserialization of untrusted data in Microsoft Office SharePoint allows an authorized attacker to execute code over a network.

Deserialization of untrusted data in Microsoft Office SharePoint allows an authorized attacker to execute code over a network.

Deserialization of untrusted data in Microsoft Office SharePoint allows an authorized attacker to execute code over a network.

Deserialization of untrusted data in Microsoft Office SharePoint allows an authorized attacker to execute code over a network.

Deserialization of untrusted data in Microsoft Office SharePoint allows an authorized attacker to execute code over a network.

Remote code execution in Adversarial Robustness Toolbox (ART) through version 1.20.1 allows unauthenticated network attackers to execute arbitrary Python code by uploading malicious PyTorch model files to pipeline-accessible object storage locations. The vulnerability stems from unsafe use of torch.load() without the weights_only=True parameter in the Kubeflow component's model loading process, enabling Pickle deserialization of arbitrary objects. With CVSS 9.8 (AV:N/AC:L/PR:N/UI:N) but only 0.06% EPSS exploitation probability (19th percentile), this represents a critical-severity issue with low observed real-world targeting, likely due to the specialized nature of ML robustness evaluation deployments. No active exploitation confirmed (not in CISA KEV) and no public exploit code identified at time of analysis.

Insecure deserialization in Optimate's neural_magic_training.py script enables remote code execution when loading PyTorch model files. The _load_model() function uses torch.load() without the weights_only=True security parameter, allowing attackers with low privileges to execute arbitrary Python code by providing malicious .pt or .pth files via the --model command-line argument. EPSS indicates low exploitation probability at 0.06% with no active exploitation confirmed.

Remote code execution in Snorkel machine learning library (≤v0.10.0) occurs when users load untrusted model files via MultitaskClassifier.load(). The vulnerability exploits insecure Python object deserialization through torch.load(), allowing attackers to embed malicious code in model weight files that executes upon loading. EPSS score of 0.06% (19th percentile) suggests low observed exploitation probability in the wild, though SSVC framework indicates total technical impact once exploited. No public exploit code or active exploitation confirmed at time of analysis, but exploitation requires only that a data scientist or ML engineer load a malicious .pkl model file.

Remote code execution in Mamba language model framework (through version 2.2.6) allows unauthenticated attackers to execute arbitrary Python code by publishing malicious models on HuggingFace Hub. When victims call MambaLMHeadModel.from_pretrained() on a weaponized model repository, insecure pickle deserialization executes attacker-controlled code in the context of the victim's process. Despite the critical CVSS 9.8 score and network attack vector requiring no authentication, EPSS probability remains extremely low (0.02%, 5th percentile), suggesting limited real-world exploitation to date. No CISA KEV listing or public POC identified at time of analysis.

Arbitrary code execution occurs in PyTorch Lightning 2.6.0 and earlier when loading malicious checkpoint files. The LightningModule.load_from_checkpoint() method deserializes untrusted Pickle data without security restrictions, allowing attackers to execute arbitrary Python code when victims open crafted .ckpt files. EPSS score of 0.06% (19th percentile) indicates low observed exploitation probability, and no public exploit code or CISA KEV listing exists at time of analysis. Attack requires local access and user interaction (opening a malicious checkpoint), limiting remote attack scenarios to social engineering or supply chain compromise.

Arbitrary code execution via torch-checkpoint-shrink.py script in ml-engineering project allows remote attackers to execute malicious Python code by providing crafted PyTorch checkpoint files. The vulnerability stems from insecure deserialization where torch.load() processes .pt files without the weights_only=True safeguard, enabling pickle-based arbitrary object instantiation. Despite a critical CVSS 9.8 score, EPSS probability is low (0.06%, 19th percentile) and no public exploit or active exploitation is confirmed, suggesting limited real-world targeting to date. SSVC assessment indicates total technical impact with automatable exploitation potential, making this a priority for organizations using ml-engineering scripts in production environments.

Arbitrary code execution in Snorkel machine learning library (≤v0.10.0) occurs when users load malicious model checkpoint files through the Trainer.load() method. The vulnerability stems from unsafe PyTorch deserialization that processes untrusted Pickle objects without the weights_only security parameter. Attackers can embed malicious Python code in model files distributed through repositories, shared datasets, or social engineering campaigns. Despite the 8.8 CVSS score indicating critical severity, EPSS scoring at 0.06% (19th percentile) suggests very low real-world exploitation probability, and no active exploitation or public proof-of-concept has been identified at time of analysis.

Remote code execution in Ludwig framework ≤0.10.4 allows unauthenticated network attackers to execute arbitrary code by supplying a malicious PyTorch model file to the ludwig serve endpoint. The vulnerability stems from unsafe deserialization in the model loading component, which uses torch.load() without the weights_only=True safety parameter. With CVSS 9.8 (critical network vector, no authentication required) but only 0.02% EPSS, this represents a high-severity issue in vulnerable deployments, though widespread exploitation has not been observed. No CISA KEV listing or public POC identified at time of analysis.

The CosyVoice project thru commit 6e01309e01bc93bbeb83bdd996b1182a81aaf11e (2025-30-21) contains an insecure deserialization vulnerability (CWE-502) in its model loading process. When loading model files (.pt) from a user-specified directory (via the --model_dir argument), the code uses torch.load() without the security-restrictive weights_only=True parameter. This allows the deserialization of arbitrary Python objects via the Pickle module. An attacker can exploit this by providing a maliciously crafted model directory containing .pt files with embedded pickle payloads. When a victim loads this directory using CosyVoice's web interface, the malicious payload is executed, leading to remote code execution on the victim's system.

Arbitrary code execution in imgaug library (versions through 0.4.0) occurs when the BackgroundAugmenter class deserializes malicious pickle payloads without validation in its multiprocessing worker method. Attackers who can influence queue data-through compromised shared queues, malicious input scripts, or social engineering-can achieve remote or local code execution depending on deployment context. CVSS 9.8 critical severity reflects network-based exploitation without authentication, though EPSS probability is low (0.02%, 6th percentile), indicating limited observed exploitation activity. No CISA KEV listing or public exploit code identified at time of analysis.

Remote code execution in Horovod distributed training framework (versions through 0.28.1) allows unauthenticated network attackers to execute arbitrary code on worker nodes by injecting malicious pickle payloads into the KVStore HTTP server. The vulnerability combines unauthenticated write access to the KVStore coordination server with unsafe deserialization using cloudpickle.loads(), enabling trivial exploitation against any reachable Horovod cluster. EPSS score of 0.12% (31st percentile) suggests low widespread exploitation probability despite critical CVSS 9.8 rating, and no active exploitation confirmed (not in CISA KEV). Public exploit development is highly feasible given the straightforward attack path and publicly documented details.

Remote code execution in Optimate's neural_magic_training.py script allows authenticated attackers to execute arbitrary code via malicious PyTorch model files. The vulnerability stems from unsafe deserialization when loading model state dictionaries without PyTorch's weights_only=True security flag, enabling pickle-based arbitrary object execution. With an EPSS score of 0.06% and no confirmed exploitation, this represents a moderate risk primarily in environments where users can upload or specify model files.

Arbitrary code execution in Snorkel library (Python) through version 0.10.0 enables remote attackers to execute code by supplying malicious pickle files to the BaseLabeler.load() method. The vulnerability stems from unsafe deserialization using pickle.load() without input validation, allowing attackers to craft serialized objects that execute arbitrary commands during deserialization. With EPSS at 6th percentile, exploitation probability remains relatively low despite the critical CVSS score, and no active exploitation (KEV) or public proof-of-concept has been identified at time of analysis.

Arbitrary code execution in Ludwig framework ≤0.10.4 occurs when attackers supply malicious pickle files to the predict() method, which deserializes untrusted data without validation using pandas.read_pickle(). Remote unauthenticated attackers can achieve full system compromise by exploiting the automatic file format detection mechanism that processes .pkl files through Python's unsafe pickle module. EPSS score of 0.06% (19th percentile) suggests low current exploitation likelihood despite the critical CVSS 9.8 rating, though no public exploit code or active exploitation has been identified at time of analysis.

An authenticated administrator who configures or tests LDAP connectivity in Sonatype Nexus Repository Manager versions 3.0.0 through 3.91.1 may be able to initiate unintended server-side connections when interacting with a malicious LDAP server.

Unsafe Python pickle deserialization in pgAdmin 4 FileBackedSessionManager allows authenticated local users with session-directory write access to execute arbitrary code as the pgAdmin process. The vulnerability arises from deserializing session files before validating their HMAC signature, enabling payload injection through crafted pickle objects. Attackers require both valid authentication and filesystem write permission to the sessions directory-achievable through misconfiguration or chaining with a separate path-traversal vulnerability. EPSS exploitation probability and KEV status not provided; no public exploit code identified at time of analysis. PostgreSQL maintainers confirmed the flaw and patched it in version 9.15 by implementing pre-deserialization HMAC validation.

CosyVoice thru commit 6e01309e01bc93bbeb83bdd996b1182a81aaf11e (2025-30-21) contains an insecure deserialization vulnerability (CWE-502) in its gRPC server component. When the server starts, it loads the speech synthesis model from a user-specified directory using torch.load() without enabling the weights_only=True security parameter. This allows the deserialization of arbitrary Python objects via the pickle module. An attacker can exploit this by providing malicious model files within a directory. When a victim starts the gRPC server pointing to this directory, arbitrary code is executed on the victim's system during server initialization.

CosyVoice thru commit 6e01309e01bc93bbeb83bdd996b1182a81aaf11e (2025-30-21) contains an insecure deserialization vulnerability (CWE-502) in its average_model.py model averaging tool. The script loads PyTorch checkpoint files (epoch_*.pt) for model averaging using torch.load() without enabling the weights_only=True security parameter. This allows the deserialization of arbitrary Python objects via the pickle module. An attacker can exploit this by providing malicious checkpoint files within a directory. When a victim uses the tool to average models from this directory, arbitrary code is executed on the victim's system.

The flash-attention training framework thru commit e724e2588cbe754beb97cf7c011b5e7e34119e62 (2025-13-04) contains an insecure deserialization vulnerability (CWE-502) in its checkpoint loading mechanism. The load_checkpoint() function in checkpoint.py and the checkpoint loading code in eval.py use torch.load() without enabling the security-restrictive weights_only=True parameter. This allows the deserialization of arbitrary Python objects via the pickle module. An attacker can exploit this by providing a maliciously crafted checkpoint file. When a victim loads this checkpoint during model warmstarting or evaluation, arbitrary code is executed on the victim's system.

CosyVoice thru commit 6e01309e01bc93bbeb83bdd996b1182a81aaf11e (2025-30-21) contains an insecure deserialization vulnerability (CWE-502) in its model loading component. The framework uses torch.load() to load model weight files (e.g., llm.pt, flow.pt, hift.pt) without enabling the security-restrictive weights_only=True parameter. This allows the deserialization of arbitrary Python objects via the pickle module. An attacker can exploit this by providing a malicious model directory containing specially crafted model files. When a victim starts the CosyVoice Web UI pointing to this directory, arbitrary code is executed on the victim's system during the model loading process.

CosyVoice thru commit 6e01309e01bc93bbeb83bdd996b1182a81aaf11e (2025-30-21) contains an insecure deserialization vulnerability (CWE-502) in its make_parquet_list.py data processing tool. The script loads PyTorch .pt files (utterance embeddings, speaker embeddings, speech tokens) using torch.load() without enabling the weights_only=True security parameter. This allows the deserialization of arbitrary Python objects via the pickle module. An attacker can exploit this by providing malicious .pt files within a data directory. When a victim processes this directory using the tool, arbitrary code is executed on the victim's system.

LangChain contains older runtime code paths that deserialize run inputs, run outputs, or other application-controlled payloads using overly broad object allowlists. These paths may call `load()` with `allowed_objects="all"`. This does not enable arbitrary Python object deserialization, but it does allow any trusted LangChain-serializable object to be revived, which is broader than these runtime paths require. As a result, attacker-supplied LangChain serialized constructor dictionaries may cause trusted runtime paths to instantiate classes with untrusted constructor arguments. Applications are exposed only when all of the following are true: 1. The application accepts untrusted structured input, such as JSON, from a user or network request. 2. The application does not validate or canonicalize that input into an inert schema before invoking LangChain. 3. Attacker-controlled nested dictionaries or lists are preserved in LangChain run inputs or outputs. 4. The application uses an affected API path that later deserializes that run data. Known affected runtime surfaces include: - `RunnableWithMessageHistory` - `astream_log()` - `astream_events(version="v1")` Related unsafe deserialization patterns may also affect applications that explicitly load serialized LangChain prompt or runnable objects from untrusted sources, including shared prompt stores, Hub artifacts with model configuration, or other application-controlled serialization stores. Applications that validate incoming requests against a fixed schema, such as coercing user input to a plain string or message-content field before invoking LangChain, are unlikely to expose this deserialization primitive. This release also fixes a related secret-marker validation bypass in the serialization and deserialization layer (`_is_lc_secret`). That issue creates an additional path by which attacker-controlled constructor dictionaries can avoid escaping during `dumps()` -> `loads()` round-trips and reach LangChain object revival logic. ## Impact An attacker who can submit untrusted structured input to an affected application, and have that structure preserved in LangChain run data, may be able to inject LangChain serialized constructor payloads such as: ```json { "lc": 1, "type": "constructor", "id": ["langchain_core", "messages", "ai", "AIMessage"], "kwargs": {"content": "attacker-controlled content"} } ``` If this payload reaches a broad `load()` call, LangChain may instantiate the referenced class instead of treating the payload as inert user data. Realistic impacts include: - Persistent chat-history poisoning when revived `AIMessage`, `HumanMessage`, or `SystemMessage` objects are stored by `RunnableWithMessageHistory`. - Prompt injection or behavior manipulation if attacker-controlled messages are later included in model context. - Instantiation of unexpected trusted LangChain objects with attacker-controlled constructor arguments. - Possible credential disclosure or server-side requests if a reachable object reads environment credentials, creates clients, or contacts attacker-controlled endpoints during initialization. - Additional prompt-template or runnable-configuration impacts in applications that separately load and execute untrusted serialized LangChain objects. ## Remediation LangChain will deprecate the affected APIs as part of this fix: - `RunnableWithMessageHistory` - `astream_log()` - `astream_events(version="v1")` These are older code paths that are no longer recommended for new applications. They were not previously marked as deprecated, but recent LangChain documentation has primarily directed users toward newer streaming and memory patterns, including the `stream` API. Applications should migrate to the currently recommended APIs rather than continue depending on these older surfaces. Separately, LangChain will update `load()` and `loads()` to tighten deserialization behavior so broad object revival is not applied implicitly to untrusted or application-controlled payloads. The older runtime surfaces listed above are being deprecated rather than preserved as supported paths for broad runtime deserialization. This release also fixes a related secret-marker validation bypass in the serialization and deserialization layer (`_is_lc_secret`). That issue creates an additional path by which attacker-controlled constructor dictionaries can avoid escaping during `dumps()` -> `loads()` round-trips and reach LangChain object revival logic. ## Guidance for `load()` and `loads()` `load()` and `loads()` should be used only with trusted LangChain manifests or serialized objects from trusted storage. Do not pass user-controlled data to `load()` or `loads()`, and do not use them as general parsers for request bodies, tool inputs, chat messages, or other attacker-controlled data. `load()` and `loads()` are beta APIs, and their behavior may change as LangChain narrows unsafe defaults. Future LangChain versions will require callers to be explicit about which objects may be revived. Users should pass a narrow `allowed_objects` value appropriate for the specific trusted manifest they are loading, rather than relying on broad defaults or `allowed_objects="all"`, which permits the full trusted LangChain serialization allowlist. ## Credits The original issue was first reported by @u-ktdi. Similar findings were reported by @dewankpant, @shrutilohani, @Moaaz-0x, @pucagit. A related `_is_lc_secret` marker bypass affecting `dumps()` -> `loads()` round-trips was reported by @yardenporat353 (and a similar report by @localhost-detect)

Remote code execution in SEPPmail Secure Email Gateway via insecure deserialization allows unauthenticated attackers to execute arbitrary code through the GINA UI interface. Versions prior to 15.0.4 deserialize untrusted data without validation, enabling attackers to send crafted serialized objects that execute upon processing. CVSS 9.2 reflects network-accessible attack with low complexity requiring only present attack conditions, though no active exploitation (KEV) or public POC has been identified at time of analysis.

PHP object injection in User Frontend plugin for WordPress versions up to 4.3.1 allows authenticated attackers with Subscriber-level access or above to achieve remote code execution via unsafe deserialization of the wpuf_files parameter during form submission. The vulnerability chains input validation failures during form processing with unconditional use of maybe_unserialize() when rendering post content, enabling attackers to inject malicious PHP objects that can execute arbitrary code, delete files, or trigger other attacks through available Property-Oriented Programming (POP) chains. Wordfence disclosed detailed code references showing the vulnerable data flow across multiple plugin files including wpuf-functions.php, FieldableTrait.php, and Frontend_Form_Ajax.php, with both trunk and version 4.2.10 code paths exhibiting the flaw.

Netgate pfSense CE 2.7.2 allows code execution by using the module installer with a backup file with a serialized PHP object containing the post_reboot_commands property. NOTE: the Supplier disputes this because this installer is only available to admins and they are intentionally allowed to execute PHP code.

LINQPad before 5.52.01 Pro edition is vulnerable to Unsafe Deserialization in LINQPad.AutoRefManager::PopulateFromCache(), leading to code execution. Rated high severity (CVSS 7.3), this vulnerability is low attack complexity. Public exploit code available and no vendor patch available.