appsec.fyi

Insecure Deserialization Resources

Post Share

A curated AppSec resource library covering XSS, SQLi, SSRF, IDOR, RCE, XXE, OSINT, and more.

Insecure Deserialization

Insecure deserialization vulnerabilities occur when applications reconstruct objects from serialized data without proper validation, potentially allowing attackers to execute arbitrary code, bypass authentication, or manipulate application logic. These vulnerabilities affect virtually every major programming language: Java (ObjectInputStream, ysoserial gadget chains), PHP (unserialize), Python (pickle, PyYAML), .NET (BinaryFormatter, Json.NET with TypeNameHandling), and Ruby (Marshal). Deserialization attacks are particularly dangerous because they often achieve remote code execution with a single crafted payload. The exploitation landscape includes gadget chain discovery, polyglot payloads that work across libraries, and attacks against message queues, caching layers, and session management systems that serialize user-controlled data. Defenses include avoiding native serialization for untrusted data, using safe alternatives like JSON, implementing allowlists for deserialized types, and integrity checking serialized objects.

The vulnerability is the type information, not the parser

Insecure deserialization is regularly explained as a parsing problem, which makes it sound like input validation would help. It would not. The issue is that serialization formats such as Java's native serialization, .NET's BinaryFormatter, PHP's unserialize and Python's pickle encode which classes to instantiate alongside the data. Reconstructing the object therefore means running that type's reconstruction logic, and in every one of these runtimes there is a hook that executes during reconstruction — readObject, __wakeup and __destruct, __reduce__. The attacker is not injecting code. They are choosing which of your existing code runs, and in what order.

That is what a gadget chain is: a sequence of ordinary, non-malicious classes already present in the application or its dependencies, arranged so that the side effects of deserializing them compose into something useful. This is why the vulnerability depends so heavily on what is on the classpath, why tools like ysoserial ship a catalogue of chains per library, and why adding a dependency can make a previously unexploitable endpoint exploitable without any change to your own code.

It is also why blocklists fail. Forbidding the classes in a published gadget chain prevents that chain and not the next one, and the history of Java deserialization mitigations is a long sequence of new chains found in libraries nobody had examined. Allowlisting the specific types an endpoint legitimately expects is the only version of type filtering that holds, and it requires knowing that set.

The durable fix is to not accept serialized objects from untrusted sources at all. A data-only format — JSON, or a schema-bound format like Protocol Buffers — carries values rather than types, so there is nothing to instantiate. Where an object graph genuinely must cross a trust boundary, signing it and verifying before deserializing moves the problem to key management, which is a better problem.

Look for these wherever serialized data travels: cookies and hidden fields, caches, message queues, session stores and inter-service calls.

Start here: the Deser learning pathAn ordered route through this library — orientation, then methodology, then research. Read the Deser guideA long-form, source-cited deep dive synthesized from every resource below. The comprehensive Deser guide on chs.usA hand-written, in-depth practitioner guide — attacks, testing, and prevention.
Date Added Link Excerpt
2026-11-10 NEW 2026rConfig deserialization (critical) news RCEThis content describes a critical deserialization vulnerability in rConfig Core versions 8.0.0 to 8.2.19 and rConfig Pro. The vulnerability allows for remote code execution, meaning attackers can potentially gain control of the affected systems without any user interaction. This is a significant security flaw that requires immediate attention and patching by users of these rConfig versions.
2026-11-10 NEW 2026Exploiting a PHP object injection in Profile Builder Pro in the era of AI intermediate AI RCEA critical PHP object injection vulnerability has been discovered in WordPress plugin "Profile Builder Pro" prior to version 3.14.5. Attackers can exploit this flaw without authentication. The discovery and exploitation involved a new POP chain, with AI significantly assisting in the process. The blog post details the methodology, including identifying real-world targets affected by this vulnerability.
2026-11-10 NEW 2026ZDI-26-850: Microsoft Qlib MLflowRecorder deserialization RCE news AI Python RCEA deserialization vulnerability in Microsoft Qlib's MLflowRecorder, rated CVSS 7.8, allows for remote code execution (RCE) if an attacker tricks a user into opening a malicious file. This flaw enables attackers to gain control of affected systems. The advisory provides guidance on how to detect and address this security risk.
2026-11-10 NEW 2026Public PoC for Hugging Face Datasets pickle deserialization RCE (CVE-2026-57845) news AI Python RCEA public proof-of-concept exploit has been released for CVE-2026-57845, a critical vulnerability in Hugging Face datasets. This flaw, rated with a CVSS score of 8.8, allows for remote code execution through insecure pickle deserialization. The vulnerability affects versions of Hugging Face datasets up to and including 5.1.0. Currently, there is no patch or fix available for this security issue.
2026-11-09 NEW 2026QA Assistants WordPress plugin PHP object injection newsWriteup detailing PHP Object Injection in the QA Assistants – Driven by data WordPress plugin (versions up to 5.3.0). This vulnerability arises from unrestricted PHP deserialization via `wrap_unserialize_static()` and improper sanitization in `serialize_escape()`. Exploitation allows unauthenticated attackers to achieve Remote Code Execution by leveraging the Guzzle FileCookieJar gadget for Arbitrary File Write, triggered via the `qahm_cron_data_manage` scheduled event.
2026-11-09 NEW 2026The classes the blocklist forgot: chaining Oracle Coherence and Xalan into RCE advanced 5 min read RCELibrary that demonstrates a novel Java deserialization RCE technique by chaining Oracle Coherence and Apache Xalan classes. This approach bypasses common blocklists that focus on known gadget chains like CommonsCollections, Spring, Groovy, and Hibernate. The exploit constructs a unique payload within a strict byte limit and uses DNS OOB callbacks for command execution, achieving arbitrary code execution on multiple production cluster nodes.
2026-11-09 NEW 2026Radius Booking deserialization of untrusted data (critical) news RCEA critical vulnerability in Radius Booking allows for deserialization of untrusted data. This means an attacker could potentially exploit the system by submitting specially crafted data that, when deserialized, leads to code execution or other malicious actions. The specific impact depends on how and where this deserialization occurs within the application. Further details on the exact exploitation vector and potential severity are not provided, but the "critical" classification highlights significant risk.
2026-11-09 NEW 2026LMDeploy CVE-2026-119706: unauthenticated RCE via pickle deserialization news 7 min read AI Python RCELibrary that fixes CVE-2026-119706 in LMDeploy, an unauthenticated RCE vulnerability stemming from unsafe pickle deserialization on the `/update_weights` endpoint. Versions 0.9.0 through 0.18.0 are affected. The fix introduces safe alternatives like safetensors, structured dictionaries, and an opt-in mechanism for pickle support via `LMDEPLOY_ALLOW_PICKLE_UPDATE_PARAMS`. → zeropath.com
2026-11-09 NEW 2026Insecure deserialization in Java, .NET, PHP, Python and Node.js intermediate 3 min read Python RCELibrary detailing insecure deserialization, a common RCE vector in Java, .NET, PHP, Python, and Node.js. It explains how to identify serialized data, construct gadget chains using tools like `ysoserial` for Java and .NET, `phpggc` for PHP, `pickle` for Python, and `node-serialize` for Node.js, and bypass deserialization protections, including specific techniques for ViewState and PHAR deserialization.
2026-11-08 NEW 2026IBM: arbitrary code execution in Qiskit SDK QPY deserialization news 2 min read AI Python RCEWriteup on CVE-2026-116499, an arbitrary code execution vulnerability in IBM's Qiskit SDK. A crafted QPY payload, when deserialized using versions prior to 13, can trigger malicious Python code execution via the `qiskit.qpy.load()` function. This CWE-502 issue has a CVSS base score of 8.4 and is addressed in Qiskit v2.6.2. → ibm.com
2026-11-08 NEW 2026MLflow SHAP load_explainer deserialization RCE news AI Python RCELibrary for identifying and mitigating the MLflow SHAP load_explainer deserialization RCE vulnerability, GHSA-j38m-78x9-f2gg. This vulnerability, also known as ZDI-CAN-30301, allows remote attackers to execute arbitrary code by exploiting improper validation of user-supplied data during SHAP explainer model loading. Exploitation requires user interaction, such as visiting a malicious page or opening a crafted file, leading to code execution within the current user's context.
2026-11-08 NEW 2026Stanza RCE via pickle deserialization in the lemmatizer intermediate 2 min read AI Python RCELibrary for detecting Remote Code Execution (RCE) vulnerabilities in the Stanza NLP toolkit. Exploitation occurs via untrusted lemmatizer checkpoints, where an attacker can supply a crafted `.pt` file containing a malicious pickle payload. This payload is deserialized by an unrestricted `pickle.loads` call within the `Trainer.load` function, leading to arbitrary code execution as the loading user. The vulnerability is triggered during `stanza.Pipeline` construction when loading a custom lemma model path. The fix involves replacing the unrestricted `pickle.loads` with Stanza's existing `RestrictedUnpickler` to sanitize deserialized data.
2026-11-08 NEW 2026MLflow PickleEvaluationArtifact deserialization RCE intermediate AI Python RCELibrary for identifying MLflow PickleEvaluationArtifact deserialization RCE, identified as GHSA-88cv-457h-5c75 and ZDI-CAN-29513. This vulnerability allows remote attackers to execute arbitrary code by exploiting the lack of proper validation during pickle file parsing, leading to deserialization of untrusted data. User interaction, such as visiting a malicious page or opening a file, is required for exploitation.
2026-11-08 NEW 2026labbench before 0.46.1 unsafe pickle deserialization in read() intermediate Python RCEVulnerability writeup detailing unsafe pickle deserialization in labbench versions prior to 0.46.1. The `read()` and `read_relational()` functions are susceptible to arbitrary code execution when processing crafted `.p` or `.pickle` files through `pandas.read_pickle`, as file type is determined solely by extension. This affects CVE-2026-120833.
2026-11-07 NEW 2026Apache MINA JavaNativeMessageDecoder unsafe deserialization RCE intermediate RCEApache MINA's JavaNativeMessageDecoder component is vulnerable to Remote Code Execution (RCE) due to unsafe deserialization. This vulnerability allows attackers to exploit the decoder's handling of serialized Java objects, potentially leading to arbitrary code execution on the server.
2026-11-07 NEW 2026Quest Foglight baseRmiServiceExporter unauthenticated deserialization RCE intermediate RCEAdvisory for GHSA-3xxj-hwx3-jcvq detailing a Deserialization of Untrusted Data Remote Code Execution vulnerability in Quest Foglight for Databases. Unauthenticated attackers can exploit the misconfiguration of the baseRmiServiceExporter bean to execute arbitrary code as SYSTEM. This flaw, tracked as ZDI-CAN-29170, stems from insufficient validation of user-supplied data during deserialization.
2026-11-07 NEW 2026OpenVox/Puppet authenticated RCE via event deserialization intermediate 2 min read RCEWriteup of CVE-2026-115731, detailing an authenticated RCE in OpenVox/Puppet servers. Attackers with client certificates can exploit a Ruby deserialization vulnerability in the `PUT /puppet/v3/report` endpoint. By crafting a specific JSON event containing a `__ptype` with a `RubyMethod` annotation, they can achieve arbitrary code execution as the `puppet` user, potentially leading to compromise of the CA signing key and further control over orchestrated hosts. This critical vulnerability also impacts Puppet Enterprise.
2026-11-07 NEW 2026ZDI-26-850 Microsoft Qlib MLflowRecorder deserialization RCE intermediate AI Python RCELibrary for analyzing deserialization vulnerabilities in Microsoft Qlib, specifically detailing ZDI-26-850, a remote code execution flaw within the MLflowRecorder class. This RCE arises from improper validation of user-supplied data, enabling an attacker to deserialize untrusted input and execute code in the context of the current user. Exploitation requires user interaction, such as visiting a malicious page or opening a crafted file. The vulnerability is patched in version v0.9.8. → zerodayinitiative.com
2026-11-07 NEW 2026Stanza RCE via pickle deserialization in the lemmatizer intermediate 3 min read AI Python RCEWriteup of CVE-2026-115878 in Stanza, a remote code execution vulnerability allowing arbitrary code execution via crafted lemma checkpoints. The flaw lies in `Trainer.load` which uses an unrestricted `pickle.loads` on gzip-compressed pickle data embedded within checkpoint files. This bypasses `torch.load`'s `weights_only=True` protection, enabling attackers to inject malicious code that runs at pipeline construction time, as demonstrated by a PoC utilizing `os.system`. The fix involves replacing the unrestricted unpickler with Stanza's `RestrictedUnpickler`. → advisories.gitlab.com
2026-11-06 NEW 2026RELATE LMS RCE via insecure Celery pickle deserialization intermediate 2 min read Python RCELibrary for identifying Remote Code Execution (RCE) in RELATE LMS via insecure Celery pickle deserialization. Attackers can execute arbitrary commands on the host server by submitting crafted pickle payloads to the message broker, exploiting Celery's default configuration and a lack of network isolation in code execution sandboxes. This vulnerability, detailed in GHSA-4mwh-mwv4-m252, allows authenticated students to achieve full RCE.
2026-11-06 NEW 2026DocsGPT unauthenticated RCE via FAISS pickle deserialization intermediate 2 min read AI Python RCEAdvisory for GHSA-m448-vp23-j393 detailing an unauthenticated Remote Code Execution vulnerability in DocsGPT. An attacker can exploit this by uploading a malicious FAISS pickle file (`index.pkl`) to the `/api/upload_index` endpoint. This bypasses authentication checks and allows for unsafe deserialization via `FAISS.load_local` with `allow_dangerous_deserialization=True`, leading to arbitrary command execution on the server.
2026-11-06 NEW 2026numcodecs Pickle codec arbitrary code execution on decode intermediate 2 min read Python RCELibrary vulnerability allowing arbitrary code execution when decoding untrusted data with the `numcodecs.pickles.Pickle` codec. The `Pickle` codec directly calls `pickle.loads` on input buffers. This vulnerability, identified in `numcodecs` versions prior to fixes, enables attackers to achieve arbitrary code execution by crafting malicious array metadata and chunk data, as simply reading affected array chunks triggers the dangerous deserialization.
2026-11-06 NEW 2026Xoscar unauthenticated cloudpickle RCE (CVE-2026-49950) intermediate AI Python RCEWriteup on CVE-2026-49950, an unauthenticated remote code execution vulnerability in Xorbits Xoscar versions prior to 0.11.2. Attackers can exploit a deserialization flaw in the actor-pool TCP channel, specifically within `SocketChannel.recv()` and `cloudpickle.loads()`, to achieve code execution as the server process, often running with root privileges on GPU nodes, by bypassing message type validation.
2026-11-06 NEW 2026Stanza RCE via pickle deserialization in lemmatizer beginner 2 min read AI Python RCELibrary vulnerability GHSA-gh9r-c94j-cmp5 describes a remote code execution flaw in the Stanza natural language processing library. Specifically, loading a crafted lemma checkpoint file (.pt) containing a malicious pickle payload can lead to arbitrary code execution due to unrestricted `pickle.loads` within the `Trainer.load` function. This bypasses the `weights_only=True` protection and can be triggered by downstream users who load untrusted lemma model paths. The fix involves deserializing the inner stream using Stanza's `RestrictedUnpickler`.
2026-11-06 NEW 2026LMCache unauthenticated pickle RCE on multiprocess transport beginner 3 min read AI Python RCELibrary advisory detailing unauthenticated remote code execution in LMCache's multiprocess ZeroMQ transport. Versions 0.3.9 through 0.5.5, and some 0.5.6 release candidates, are affected. The vulnerability arises from unauthenticated pickle deserialization of DeviceIPCWrapper objects, allowing attackers to execute arbitrary code as the LMCache process user. This exploit targets deployments where the ZeroMQ transport binds to a routable address. The fix, implemented in version 0.5.6, replaces pickle with validated MessagePack descriptors.
2026-11-02 2026Insecure Deserialization: How to Turn Trusted Data Into a Reverse Shell intermediateThis content details how insecure deserialization, a vulnerability often found in PHP and Java web applications, can be exploited. Attackers can leverage this flaw to turn seemingly trusted data into a reverse shell, gaining control of the server. The guide offers a hands-on lab walkthrough to demonstrate these attacks, highlighting the critical importance of understanding and mitigating this security risk.
2026-11-02 2026ZDI-26-797 Meta FAIRChem pickle deserialization news AI PythonThis advisory, ZDI-26-797, details a pickle deserialization vulnerability in Meta FAIRChem. The flaw allows an attacker to execute arbitrary code by crafting a malicious pickle file. This could lead to significant system compromise. → zerodayinitiative.com
2026-11-02 2026ZDI-26-845 MLflow deserialization of untrusted data news AI Python RCEAdvisory ZDI-26-845 details a remote code execution vulnerability in MLflow's `_load_model` function, stemming from insufficient validation allowing deserialization of untrusted data. Exploitation requires user interaction, typically by visiting a malicious page or opening a crafted file, enabling attackers to execute code within the context of the current user. This flaw was fixed in MLflow version 3.15.0. → zerodayinitiative.com
2026-11-01 2026IBM webMethods Integration insecure deserialization (CVE-2026-12118) news RCEIBM WebMethods Integration is vulnerable to insecure deserialization due to CVE-2026-12118. This flaw, identified by CPAI-2026-14682, allows for potential exploitation through improperly handled serialized data. The severity of this vulnerability lies in its ability to compromise system integrity and security. Further details and mitigation strategies are available via the provided CVE and CPAI identifiers.
2026-11-01 2026Seroval: deserialization invokes attacker-chosen callables (CVE-2026-116580) intermediate 1 min read Supply ChainLibrary patch for Seroval addresses CVE-2026-116580, a deserialization vulnerability where attacker-chosen callables could be invoked. The flaw, a variant of GHSA-mv8w-475r-vwqw, allowed crafted payloads to cause untrusted input passed to `fromJSON` to execute arbitrary functions via the `on` method. Versions up to 1.6.7 are affected; upgrade to 1.6.8 to fix.
2026-11-01 2026Mautic insecure deserialization in segment filters and custom fields intermediate 1 min read RCEAdvisory detailing GHSA-5r63-8h35-j3mw, an insecure deserialization vulnerability in Mautic's segment filters and custom fields. Exploitation allows arbitrary PHP object instantiation, potentially leading to remote code execution, data compromise, or denial of service. The vulnerability affects Mautic versions prior to 7.2.0, 6.0.10, and 5.2.12. Workarounds involve restricting administrative access and database write permissions.
2026-11-01 2026ZDI-26-839 MLflow PickleEvaluationArtifact deserialization RCE news AI Python RCEAdvisory ZDI-26-839 details a remote code execution vulnerability in MLflow due to deserialization of untrusted pickle data. Affected installations allow remote attackers to execute arbitrary code in the context of the current process if a target user visits a malicious page or opens a malicious file. This flaw was fixed in MLflow 3.13.0. → zerodayinitiative.com
2026-10-31 2026GHSL-2026-200: Untrusted Seroval payloads intermediate 3 min readAdvisory for GHSL-2026-200 detailing a type confusion vulnerability in the `seroval` serialization library. Failing to validate the type of the "buffer" child in typed-array and data-view deserialization allows attackers to trigger denial of service by causing excessive memory allocation and CPU burn with small payloads, or even invoke registered functions. This impacts servers deserializing untrusted JSON bodies, such as those used in TanStack Start server functions, and is a variant of CVE-2026-59940. → securitylab.github.com
2026-10-31 2026Gradle: deserialization of untrusted Java objects before authentication intermediate 2 min read AuthN RCEAdvisory for GHSA-xwqc-3h47-hg64 details a remote code execution vulnerability in Gradle's client-daemon communication. Attackers can exploit this by deserializing untrusted Java objects before authentication, allowing for code execution within the daemon JVM with the privileges of the build user. Gradle has patched this by verifying the secret token before deserialization. Workarounds include avoiding shared build hosts and ensuring only trusted users have local access. This advisory references Gradle's own ADR-0002 and JEP 290.
2026-10-31 2026Serena RCE via insecure pickle deserialization of LSP symbol cache advanced 5 min read AI Python RCE Supply ChainLibrary for detecting RCE via insecure pickle deserialization of the LSP symbol cache in Serena. Versions prior to 1.7.1 are affected. A malicious repository can trigger arbitrary code execution as the user running Serena simply by being opened or activated, exploiting the unconditional loading of pickled cache files located within the project directory. This vulnerability, identified as CWE-502, bypasses existing trust mechanisms and allows attackers to steal credentials, install persistence, or take over the host.
2026-10-31 2026Xoscar unauthenticated RCE through actor message deserialization advanced 4 min read AI Python RCELibrary for securing Xoscar applications against unauthenticated RCE. Versions v0.1.0 through v0.11.2, and v0.12.0 with an unset `XOSCAR_AUTH_KEY`, are vulnerable to arbitrary code execution via actor message deserialization using `cloudpickle.loads`. Upgrading to v0.12.0 and configuring a strong, shared `XOSCAR_AUTH_KEY` is required to authenticate socket and UCX messages before deserialization, mitigating the vulnerability and related control-plane exposures like `get_config` and `stop`.
2026-10-31 2026ZDI-26-816 Ultralytics load_dataset_cache_file deserialization RCE news AI Python RCEAdvisory ZDI-26-816 details a deserialization of untrusted data vulnerability in Ultralytics' `load_dataset_cache_file` function. This flaw permits remote code execution on affected installations, requiring user interaction through visiting a malicious page or opening a malicious file. The vulnerability stems from insufficient validation of user-supplied data, allowing attackers to execute code within the context of the current user. Mitigation involves restricting interaction with the product. → zerodayinitiative.com
2026-10-26 2026Broccoli2Shell: Brotli decoder RCE in Node.js, Python and more intermediate 6 min read RCE Supply ChainWriteup detailing a Brotli decoder RCE vulnerability affecting Node.js, Python, and other software. Exploitation occurs when code decompresses untrusted Brotli data, impacting HTTP clients, request body parsers like Express and body-parser, and direct decoder calls. Vulnerable versions of the C decoder (Brotli 1.0.3-1.2.0) and its bindings are exposed. The fix involves upgrading Node.js, Python Brotli packages, or system libraries to patched versions.
2026-10-13 2026I Found a $90k Remote Code Execution Vulnerability in Google news 4 min read RCEWriteup detailing a $90,000 RCE vulnerability in Google's developer sites, specifically leveraging a pickle deserialization flaw in Flask session cookies. The vulnerability allowed an attacker to forge session cookies by using the exposed OAuth client secret as the Flask SECRET_KEY. This enabled arbitrary code execution via `pickle.loads` on the server-side callback, as demonstrated by using `time.sleep()` to control response times.
2026-09-26 2026Master Key Included: Detecting SolarWinds ARM CVE-2026-28326 intermediate 11 min read RCETool for detecting CVE-2026-28326, an unauthenticated RCE in SolarWinds Access Rights Manager. This vulnerability, rated 8.8 CVSS, allows attackers to achieve SYSTEM-level execution by exploiting a hardcoded client-authentication secret on TCP port 55555, leading to .NET deserialization. The tool uses two safe requests to identify vulnerable assets, which are exploitable in builds prior to 2026.2.1.7 and expose sensitive identity governance data. → bishopfox.com
2026-09-22 2026Implant Encryption via the Dump Encoding Library intermediate 4 min readLibrary for abusing the Windows Error Reporting Dump Encoding Library (WerEnc.dll) to encrypt malware implants. Threat actors can leverage AES-256 encryption provided by this Microsoft-signed DLL, offering enhanced evasion capabilities against EDR systems by reducing the need for custom crypto code and obscuring analysis with keys stored in controlled infrastructure. Proof-of-concept tools like `werenc-byok.exe` and `werenc-rt.exe` demonstrate key generation, encryption of executables (e.g., `calc.bin`), and decryption of staged implants, even fetched via HTTP. Detection focuses on monitoring arbitrary processes loading `WerEnc.dll` and associated file creation events. → ipurple.team
2026-09-07 2026From Padding Oracle to Shell: Unauthenticated RCE in Telerik UI for ASP.NET AJAX advanced 48 min read RCELibrary for unauthenticated RCE in Telerik UI for ASP.NET AJAX, exploiting an AES-CBC padding oracle chained with other vulnerabilities. Affects versions 2010.1.309 through 2026.2.519, with version 2026.2.708 providing a fix. Exploitation requires a reachable RadAsyncUpload control with a custom encryption key.
2026-08-27 2026Ruby Marshal Kick-off Gadgets - elttam intermediate 15 min readLibrary detailing nine ways `Marshal.load` on Ruby 4.0.6 can grant attackers code execution. It focuses on "kick-off gadgets," the initial method calls `Marshal.load` implicitly triggers, distinct from intermediate deserialization gadgets. The article categorizes these into documented entrypoints like `marshal_load`, `_load`, and `_load_data`, and unexpected ones such as `hash` and `eql?`, which are triggered through fundamental Ruby features and are less likely to be patched by maintainers. → elttam.com
2026-08-14 2026Ruby 4.0 Universal RCE Deserialization Gadget Chain - elttam intermediate 12 min read RCELibrary for Ruby 4.0 Universal RCE Deserialization, building on prior research into universal gadget chains. This new chain exploits `Marshal.load` and uses gadgets like `Gem::Specification.load` and `Gem::StubSpecification` to achieve command execution by controlling file content and execution paths, re-using `call_url_and_create_folder` for file writing and `Time` deserialization as a robust trigger for the download and write process. → elttam.com
2026-07-22 2026Leaking internal headers in Flask Ninja with deserialization intermediate 5 min read PythonLibrary for Python's Flask Ninja enabling API security. It details a deserialization vulnerability in Flask Ninja's HttpBearer class that can lead to internal header leaks, specifically mentioning the `pickle` gadget and its use with Pydantic models. The writeup also outlines developer-side mitigations like avoiding reflection of rejected credentials and not deserializing untrusted input.
2026-06-10 2026Security Advisory: Critical RCE Vulnerabilities in React Server Components (CVE-2025-55182) news 4 min read RCEAdvisory detailing CVE-2025-55182, a critical RCE vulnerability in React Server Components. This flaw, present in React 19 and Next.js 15/16, arises from unsafe deserialization of attacker-controlled data within the "Flight" protocol, enabling unauthenticated remote code execution via crafted HTTP requests. The vulnerability affects any framework embedding the RSC implementation, including Vite, Parcel, and Redwood. Patches are available for React and Next.js, and immediate upgrades are recommended to prevent server compromise. → snyk.io
2026-04-22 2026picoCTF Super Serial Writeup: PHP Object Injection Explained Clearly beginnerpicoCTF Super Serial Writeup: PHP Object Injection Explained Clearly
2026-04-22 2026Deep Dive into Fastjson Deserialization Vulnerabilities advancedDeep Dive into Fastjson Deserialization Vulnerabilities
2026-04-22 2026CVE-2025-24813 PoC: Apache Tomcat Java Deserialization newsProof-of-concept exploit for CVE-2025-24813, a Java deserialization vulnerability in Apache Tomcat. This tool, built with Docker, automates the generation and delivery of malicious serialized Java payloads using `ysoserial` or custom Java, targeting session files to trigger deserialization. It supports custom commands and can disable SSL verification for testing.
2026-04-22 2026WSUS Deserialization Exploit in the Wild (CVE-2025-59287) news 4 min readAnalysis of CVE-2025-59287, a WSUS deserialization vulnerability, details an in-the-wild exploitation targeting internet-exposed Windows Server Update Services. The exploit bypasses initial proof-of-concept limitations by delivering a custom .NET payload via a ysoserial.net gadget chain, enabling arbitrary command execution through an embedded `cmd.exe` process launched by `w3wp.exe`. This attack was observed via EDR telemetry, specifically the execution of `whoami.exe` with `w3wp.exe` as its parent, and confirmed by log analysis revealing deserialization errors.

Frequently Asked Questions

What makes deserialization vulnerabilities so dangerous?
Deserialization vulnerabilities often lead directly to remote code execution (RCE) because the deserialization process can trigger arbitrary method calls through gadget chains — sequences of existing classes whose methods chain together to execute attacker-controlled commands. A single crafted payload can compromise an entire server.
What is a gadget chain?
A gadget chain is a sequence of existing classes and methods in an application's classpath that, when triggered during deserialization, produce a dangerous side effect like code execution. Tools like ysoserial (Java), phpggc (PHP), and peas (Python) generate payloads for known gadget chains in popular libraries and frameworks.
How do you prevent insecure deserialization?
Never deserialize untrusted data using native serialization formats (Java ObjectInputStream, Python pickle, PHP unserialize). Use safe data formats like JSON for data exchange. If native serialization is required, implement strict type allowlists, integrity checks (HMAC signatures), and consider using look-ahead deserialization that validates types before instantiation.

Weekly AppSec Digest

Get new resources delivered every Monday.