CVE-2026-35626: OpenClaw unauthenticated DoS

CISO Take

OpenClaw before 2026.3.22 buffers the entire voice call webhook request body into memory before validating the provider signature, letting any unauthenticated attacker exhaust server resources with oversized or rapid-fire requests. The attack requires zero authentication, zero user interaction, and low complexity — EPSS places this in the top 62% most likely to be exploited, and the voice webhook endpoint is typically internet-exposed by design for telephony integrations. While not yet in CISA KEV and limited to 4 tracked downstream dependents, any production AI agent deployment using OpenClaw for voice orchestration faces real availability risk from a trivially launched flood. Patch to 2026.3.22 or later immediately; as an interim control, enforce maximum content-length limits and per-IP rate limiting on the voice webhook path at your WAF or reverse proxy before requests reach OpenClaw.

Sources: NVD EPSS GitHub Advisory ATLAS VulnCheck

What is the risk?

Medium CVSS (5.3) understates operational risk for live AI agent deployments: no authentication, no user interaction, and low complexity means any attacker can trigger this without specialized knowledge. The EPSS score of 0.00494 placing it in the top 62% reflects a realistic and near-term exploitation probability. Impact is bounded to availability (A:L in CVSS), but for AI agents handling real-time voice interactions, even partial memory exhaustion cascades into service outages that halt agentic pipelines entirely. The 396 CVEs already attributed to the same package ecosystem signals persistent security debt and an active researcher focus that elevates discovery and exploitation risk.

How does the attack unfold?

Reconnaissance

Attacker identifies an internet-exposed OpenClaw deployment and its voice call webhook endpoint path via passive OSINT on public configurations or lightweight HTTP probing.

AML.T0006

Exploitation

Attacker floods the webhook endpoint with concurrent POST requests carrying large bodies; OpenClaw buffers each request fully in memory before reaching the provider signature validation check.

AML.T0049

Impact

Server memory and CPU are exhausted by sustained unauthenticated buffering, causing the AI voice agent to become unavailable for legitimate telephony-triggered workflows and halting all downstream agentic pipelines that depend on it.

AML.T0029

Reconnaissance

Attacker identifies an internet-exposed OpenClaw deployment and its voice call webhook endpoint path via passive OSINT on public configurations or lightweight HTTP probing.

AML.T0006

Exploitation

Attacker floods the webhook endpoint with concurrent POST requests carrying large bodies; OpenClaw buffers each request fully in memory before reaching the provider signature validation check.

AML.T0049

Impact

Server memory and CPU are exhausted by sustained unauthenticated buffering, causing the AI voice agent to become unavailable for legitimate telephony-triggered workflows and halting all downstream agentic pipelines that depend on it.

AML.T0029

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
OpenClaw	pip	—	No patch
4 dependents 37% patched ~3d to patch Full package profile →

Do you use OpenClaw? You're affected.

How severe is it?

CVSS 3.1

5.3 / 10

EPSS

0.5%

chance of exploitation in 30 days

Higher than 40% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

No known exploitation

Sophistication

Trivial

What is the attack surface?

AV Network

AC Low

PR None

UI None

S Unchanged

C None

I None

A Low

What should I do?

5 steps

Patch: Upgrade OpenClaw to 2026.3.22 or later (commits 630f1479 and 651dc745 on GitHub).
Immediate workaround: Configure your reverse proxy or WAF to enforce a strict maximum request body size (e.g., 64 KB) on the voice call webhook path before requests reach OpenClaw's handler.
Add rate limiting per source IP on the webhook endpoint to cap burst volume.
Detection: Monitor for anomalous spikes in memory or CPU tied to the webhook handler process; alert on requests with unusually large Content-Length headers targeting the webhook path.
Audit: Review all other webhook endpoints in your OpenClaw deployment for the same pre-validation buffering pattern, and validate that signature checks occur before any body is read into memory.

What does CISA's SSVC say?

Decision Track

Exploitation none

Automatable Yes

Technical Impact partial

Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.

How is it classified?

DoS Agent AML.T0006 - Active Scanning AML.T0029 - Denial of AI Service AML.T0034 - Cost Harvesting AML.T0049 - Exploit Public-Facing Application

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 15 - Accuracy, robustness and cybersecurity

ISO 42001

6.1.2 - AI risk assessment

NIST AI RMF

MANAGE 2.2 - Sustain deployed AI system value

OWASP LLM Top 10

LLM10 - Model Denial of Service

Frequently Asked Questions

What is CVE-2026-35626?

OpenClaw before 2026.3.22 buffers the entire voice call webhook request body into memory before validating the provider signature, letting any unauthenticated attacker exhaust server resources with oversized or rapid-fire requests. The attack requires zero authentication, zero user interaction, and low complexity — EPSS places this in the top 62% most likely to be exploited, and the voice webhook endpoint is typically internet-exposed by design for telephony integrations. While not yet in CISA KEV and limited to 4 tracked downstream dependents, any production AI agent deployment using OpenClaw for voice orchestration faces real availability risk from a trivially launched flood. Patch to 2026.3.22 or later immediately; as an interim control, enforce maximum content-length limits and per-IP rate limiting on the voice webhook path at your WAF or reverse proxy before requests reach OpenClaw.

Is CVE-2026-35626 actively exploited?

No confirmed active exploitation of CVE-2026-35626 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-35626?

1. Patch: Upgrade OpenClaw to 2026.3.22 or later (commits 630f1479 and 651dc745 on GitHub). 2. Immediate workaround: Configure your reverse proxy or WAF to enforce a strict maximum request body size (e.g., 64 KB) on the voice call webhook path before requests reach OpenClaw's handler. 3. Add rate limiting per source IP on the webhook endpoint to cap burst volume. 4. Detection: Monitor for anomalous spikes in memory or CPU tied to the webhook handler process; alert on requests with unusually large Content-Length headers targeting the webhook path. 5. Audit: Review all other webhook endpoints in your OpenClaw deployment for the same pre-validation buffering pattern, and validate that signature checks occur before any body is read into memory.

What systems are affected by CVE-2026-35626?

This vulnerability affects the following AI/ML architecture patterns: AI agent frameworks, Voice-integrated agentic systems, Webhook-triggered automation pipelines, Telephony-driven LLM orchestration.

What is the CVSS score for CVE-2026-35626?

CVE-2026-35626 has a CVSS v3.1 base score of 5.3 (MEDIUM). The EPSS exploitation probability is 0.49%.

What is the AI security impact?

Affected AI Architectures

AI agent frameworksVoice-integrated agentic systemsWebhook-triggered automation pipelinesTelephony-driven LLM orchestration

MITRE ATLAS Techniques

AML.T0006 Active Scanning

AML.T0029 Denial of AI Service

AML.T0034 Cost Harvesting

AML.T0049 Exploit Public-Facing Application

Compliance Controls Affected

EU AI Act: Article 15

ISO 42001: 6.1.2

NIST AI RMF: MANAGE 2.2

OWASP LLM Top 10: LLM10

What are the technical details?

Original Advisory

OpenClaw before 2026.3.22 contains an unauthenticated resource exhaustion vulnerability in voice call webhook handling that buffers request bodies before provider signature checks. Attackers can send large or malicious webhook requests to exhaust server resources without authentication by bypassing signature validation.

Exploitation Scenario

An attacker targeting an organization's AI-powered voice agent discovers the OpenClaw webhook endpoint — typically a predictable path such as /webhook/voice — via passive OSINT on job postings, public configuration files, or lightweight HTTP probing. They write a simple script that fires concurrent POST requests with multi-megabyte bodies to the endpoint. OpenClaw's handler allocates heap memory to buffer each request in full before reaching the HMAC signature check, at which point it rejects unsigned requests and frees memory — but too slowly under sustained concurrent load. By keeping a large number of in-flight oversized requests active simultaneously, the attacker exhausts the server's available memory and CPU, causing the AI voice agent to stop responding to legitimate calls and potentially crashing the service process entirely, taking down all downstream agentic workflows that depend on it.

Weaknesses (CWE)

CWE-405 Asymmetric Resource Consumption (Amplification)

CWE-405 — Asymmetric Resource Consumption (Amplification): The product does not properly control situations in which an adversary can cause the product to consume or produce excessive resources without requiring the adversary to invest equivalent work or otherwise prove authorization, i.e., the adversary's influence is "asymmetric."

[Architecture and Design] An application must make resources available to a client commensurate with the client's access level.
[Architecture and Design] An application must, at all times, keep track of allocated resources and meter their usage appropriately.

Source: MITRE CWE corpus.