CVE-2024-9070 — CRITICAL (CVSS 9.8) AI Security Vulnerability

CISO Take

BentoML's runner server deserializes untrusted data without validation when the args-number parameter exceeds 1, enabling unauthenticated remote code execution by any network-accessible attacker. Any ML serving infrastructure running BentoML ≤1.3.4.post1 is at critical risk of full server compromise, including exfiltration of model weights, credentials, and training data. Immediately isolate runner server endpoints from untrusted networks and prioritize patching to a confirmed fixed version.

Risk Assessment

CVSS 9.8 with a fully unauthenticated, network-accessible, zero-interaction attack path makes this trivially exploitable by any actor with network reachability to the runner server. While EPSS (0.25%) indicates no confirmed active exploitation at time of publication, the simplicity of the trigger—a single parameter manipulation with a crafted payload—means weaponization is low-effort and public PoC development is likely imminent. ML inference infrastructure is a high-value target for credential and IP theft, and runner servers are frequently misconfigured as internally reachable from broader cloud or Kubernetes environments.

Affected Systems

Package	Ecosystem	Vulnerable Range	Patched
bentoml	pip	<= 1.4.5	No patch
8.6K OpenSSF 6.3 22 dependents Pushed 9d ago 50% patched ~14d to patch Full package profile →

Do you use bentoml? You're affected.

Severity & Risk

CVSS 3.1

9.8 / 10

EPSS

0.4%

chance of exploitation in 30 days

Higher than 61% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

Exploit Available

Exploitation: MEDIUM

Sophistication

Trivial

Exploitation Confidence

medium

○ CISA SSVC: Public PoC

Composite signal derived from CISA KEV, CISA SSVC, EPSS, trickest/cve, and Nuclei templates.

Attack Surface

AV Network

AC Low

PR None

UI None

S Unchanged

C High

I High

A High

Recommended Action

5 steps

IMMEDIATE

Firewall or network-policy restrict all BentoML runner server ports to trusted internal services only—do not expose to public internet or untrusted segments.
PATCH

Upgrade BentoML to the latest available release; verify fix inclusion by reviewing runner_app.py changes in the v1.4.5+ GitHub history. Confirm patched: N/A in current data—check https://github.com/bentoml/BentoML/releases for official guidance.
DETECT

Review HTTP access logs on runner servers for requests with args-number > 1 originating from unexpected sources; alert on anomalous child process spawning from runner processes.
AUDIT

Enumerate all BentoML deployments in your environment including Kubernetes pods, Docker Compose stacks, and cloud VMs; check network exposure for each.
COMPENSATING CONTROL

If patching is blocked, deploy a reverse proxy or API gateway in front of runner endpoints with strict request validation rejecting args-number > 1.

CISA SSVC Assessment

Decision Attend

Exploitation poc

Automatable Yes

Technical Impact total

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

Classification

Code Execution Supply Chain Framework Inference AML.T0010.001 - AI Software AML.T0049 - Exploit Public-Facing Application AML.T0072 - Reverse Shell

Compliance Impact

This CVE is relevant to:

EU AI Act

Article 15 - Accuracy, robustness and cybersecurity

ISO 42001

A.6.1.2 - AI risk assessment A.9.3 - AI system security

NIST AI RMF

MANAGE 2.2 - Treatments, responses, and recovery actions are taken

OWASP LLM Top 10

LLM05:2025 - Supply Chain Vulnerabilities

Frequently Asked Questions

What is CVE-2024-9070?

BentoML's runner server deserializes untrusted data without validation when the args-number parameter exceeds 1, enabling unauthenticated remote code execution by any network-accessible attacker. Any ML serving infrastructure running BentoML ≤1.3.4.post1 is at critical risk of full server compromise, including exfiltration of model weights, credentials, and training data. Immediately isolate runner server endpoints from untrusted networks and prioritize patching to a confirmed fixed version.

Is CVE-2024-9070 actively exploited?

No confirmed active exploitation of CVE-2024-9070 has been reported, but organizations should still patch proactively.

How to fix CVE-2024-9070?

1. IMMEDIATE: Firewall or network-policy restrict all BentoML runner server ports to trusted internal services only—do not expose to public internet or untrusted segments. 2. PATCH: Upgrade BentoML to the latest available release; verify fix inclusion by reviewing runner_app.py changes in the v1.4.5+ GitHub history. Confirm patched: N/A in current data—check https://github.com/bentoml/BentoML/releases for official guidance. 3. DETECT: Review HTTP access logs on runner servers for requests with args-number > 1 originating from unexpected sources; alert on anomalous child process spawning from runner processes. 4. AUDIT: Enumerate all BentoML deployments in your environment including Kubernetes pods, Docker Compose stacks, and cloud VMs; check network exposure for each. 5. COMPENSATING CONTROL: If patching is blocked, deploy a reverse proxy or API gateway in front of runner endpoints with strict request validation rejecting args-number > 1.

What systems are affected by CVE-2024-9070?

This vulnerability affects the following AI/ML architecture patterns: model serving, ML inference pipelines, MLOps platforms, containerized AI workloads.

What is the CVSS score for CVE-2024-9070?

CVE-2024-9070 has a CVSS v3.1 base score of 9.8 (CRITICAL). The EPSS exploitation probability is 0.41%.

Technical Details

NVD Description

A deserialization vulnerability exists in BentoML's runner server in bentoml/bentoml versions <=1.3.4.post1. By setting specific parameters, an attacker can execute unauthorized arbitrary code on the server, causing severe harm. The vulnerability is triggered when the args-number parameter is greater than 1, leading to automatic deserialization and arbitrary code execution.

Exploitation Scenario

An adversary scans internal ML infrastructure or a misconfigured cloud environment and identifies an exposed BentoML runner server HTTP endpoint. They craft a POST request to the runner API, setting args-number to 2 and embedding a malicious Python pickle payload in the request body. The server automatically deserializes the payload without any validation, executing the adversary's code—such as dropping a reverse shell, exfiltrating model weights and API keys from the serving environment, or establishing persistence via a cron job or systemd unit. No credentials, authentication tokens, AI/ML domain knowledge, or special tooling required. The full attack chain executes in seconds and is indistinguishable from legitimate runner traffic in standard HTTP logs.