CVE-2026-53872: picklescan arbitrary file read

CISO Take

picklescan, the de-facto AI/ML security tool for detecting malicious pickle files in model pipelines, contains a vulnerability where it can itself be weaponized: a crafted pickle payload chains io.FileIO with urllib.request.urlopen to exfiltrate arbitrary server files — bypassing the RCE-focused blocklist that is picklescan's own defense mechanism. This creates a false sense of security for teams that rely on picklescan to validate untrusted model checkpoints or third-party artifacts before ingestion. With no authentication required, no user interaction, and a network-accessible attack vector (CVSS 7.5), any CI/CD pipeline or MLOps workflow running picklescan ≤0.0.34 on untrusted artifacts is exposed to credential theft and configuration exfiltration from the scanning host. Upgrade to picklescan 0.0.35 immediately and rotate credentials on any host that scanned untrusted pickle files under the vulnerable version.

Sources: NVD GitHub Advisory ATLAS

What is the risk?

HIGH risk for AI/ML teams with automated pickle scanning in their artifact intake pipelines. The exploit is particularly dangerous because the attack surface is the security control itself — teams running picklescan believe they are protected while their scanner is the vector. CVSS 7.5 with AV:N/AC:L/PR:N/UI:N means the bar for exploitation is low: no special position on the network, no credentials, no victim interaction required beyond the automated scan job running. The file read primitive exposes API keys, SSH credentials, .env files, and model secrets stored on the scanning host. Although direct downstream dependents are low (3), picklescan is commonly embedded in model intake workflows processing community artifacts from registries like Hugging Face — high-value environments for credential theft.

How does the attack unfold?

Deliver Malicious Artifact

Attacker publishes a crafted pickle file disguised as a legitimate model checkpoint to a public repository or submits it to a victim's model intake pipeline.

AML.T0011.000

Trigger Vulnerable Scanner

picklescan ≤0.0.34 deserializes the crafted payload; the io.FileIO + urllib.request.urlopen chain executes while bypassing RCE-focused blocklist detection.

AML.T0049

Local File Read

The deserialized payload reads arbitrary files from the scanning host's filesystem — credential files, .env secrets, SSH keys, or cloud provider tokens.

AML.T0037

Exfiltrate Sensitive Data

File contents are transmitted via HTTP to an attacker-controlled server using urllib.request.urlopen, completing credential theft with no code execution and no scanner alert.

AML.T0025

Deliver Malicious Artifact

Attacker publishes a crafted pickle file disguised as a legitimate model checkpoint to a public repository or submits it to a victim's model intake pipeline.

AML.T0011.000

Trigger Vulnerable Scanner

picklescan ≤0.0.34 deserializes the crafted payload; the io.FileIO + urllib.request.urlopen chain executes while bypassing RCE-focused blocklist detection.

AML.T0049

Local File Read

The deserialized payload reads arbitrary files from the scanning host's filesystem — credential files, .env secrets, SSH keys, or cloud provider tokens.

AML.T0037

Exfiltrate Sensitive Data

File contents are transmitted via HTTP to an attacker-controlled server using urllib.request.urlopen, completing credential theft with no code execution and no scanner alert.

AML.T0025

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
picklescan	pip	—	No patch
419 3 dependents Pushed 9d ago 53% patched ~12d to patch Full package profile →

Do you use picklescan? You're affected.

How severe is it?

CVSS 3.1

7.5 / 10

EPSS

0.5%

chance of exploitation in 30 days

Higher than 41% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

Exploit Available

Exploitation: MEDIUM

Sophistication

Moderate

Exploitation Confidence

medium

○ CISA SSVC: Public PoC

Composite signal derived from CISA KEV, VulnCheck KEV, CISA SSVC, EPSS, Metasploit, Exploit-DB, trickest/cve, Nuclei templates, and inthewild.io exploitation reports.

What is the attack surface?

AV Network

AC Low

PR None

UI None

S Unchanged

C High

I None

A None

What should I do?

6 steps

Upgrade picklescan to 0.0.35 — this is the confirmed patched release.
If patching is not immediately possible, suspend automated pickle scanning and perform manual inspection in fully air-gapped, network-isolated environments.
Audit outbound HTTP/S traffic logs from scanning hosts for unexpected urllib or requests calls to external IPs, which would indicate active exploitation.
Rotate all credentials (API keys, cloud IAM tokens, SSH keys, .env secrets) stored on hosts that ran vulnerable picklescan against untrusted artifacts.
Apply network egress controls on scanning hosts to block unexpected outbound connections regardless of patching status.
Review CI/CD pipeline configurations to ensure scanning jobs run with minimal required permissions and no access to production secrets.

What does CISA's SSVC say?

Decision Track*

Exploitation poc

Automatable Yes

Technical Impact partial

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

How is it classified?

Data Extraction Supply Chain Framework Model AML.T0010.001 - AI Software AML.T0011.000 - Unsafe AI Artifacts AML.T0025 - Exfiltration via Cyber Means AML.T0037 - Data from Local System AML.T0049 - Exploit Public-Facing Application

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 9 - Risk management system

ISO 42001

8.4 - AI system lifecycle management

NIST AI RMF

MANAGE 2.2 - Mechanisms for AI risk response

OWASP LLM Top 10

LLM05:2025 - Supply Chain Vulnerabilities

Frequently Asked Questions

What is CVE-2026-53872?

picklescan, the de-facto AI/ML security tool for detecting malicious pickle files in model pipelines, contains a vulnerability where it can itself be weaponized: a crafted pickle payload chains io.FileIO with urllib.request.urlopen to exfiltrate arbitrary server files — bypassing the RCE-focused blocklist that is picklescan's own defense mechanism. This creates a false sense of security for teams that rely on picklescan to validate untrusted model checkpoints or third-party artifacts before ingestion. With no authentication required, no user interaction, and a network-accessible attack vector (CVSS 7.5), any CI/CD pipeline or MLOps workflow running picklescan ≤0.0.34 on untrusted artifacts is exposed to credential theft and configuration exfiltration from the scanning host. Upgrade to picklescan 0.0.35 immediately and rotate credentials on any host that scanned untrusted pickle files under the vulnerable version.

Is CVE-2026-53872 actively exploited?

No confirmed active exploitation of CVE-2026-53872 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-53872?

1. Upgrade picklescan to 0.0.35 — this is the confirmed patched release. 2. If patching is not immediately possible, suspend automated pickle scanning and perform manual inspection in fully air-gapped, network-isolated environments. 3. Audit outbound HTTP/S traffic logs from scanning hosts for unexpected urllib or requests calls to external IPs, which would indicate active exploitation. 4. Rotate all credentials (API keys, cloud IAM tokens, SSH keys, .env secrets) stored on hosts that ran vulnerable picklescan against untrusted artifacts. 5. Apply network egress controls on scanning hosts to block unexpected outbound connections regardless of patching status. 6. Review CI/CD pipeline configurations to ensure scanning jobs run with minimal required permissions and no access to production secrets.

What systems are affected by CVE-2026-53872?

This vulnerability affects the following AI/ML architecture patterns: Training pipelines, Model artifact intake workflows, CI/CD security scanning, Model serving with pre-ingestion scanning, MLOps artifact repositories.

What is the CVSS score for CVE-2026-53872?

CVE-2026-53872 has a CVSS v3.1 base score of 7.5 (HIGH). The EPSS exploitation probability is 0.51%.

What is the AI security impact?

Affected AI Architectures

Training pipelinesModel artifact intake workflowsCI/CD security scanningModel serving with pre-ingestion scanningMLOps artifact repositories

MITRE ATLAS Techniques

AML.T0010.001 AI Software

AML.T0011.000 Unsafe AI Artifacts

AML.T0025 Exfiltration via Cyber Means

AML.T0037 Data from Local System

AML.T0049 Exploit Public-Facing Application

Compliance Controls Affected

EU AI Act: Article 9

ISO 42001: 8.4

NIST AI RMF: MANAGE 2.2

OWASP LLM Top 10: LLM05:2025

What are the technical details?

Original Advisory

picklescan before 0.0.35 contains an unsafe pickle deserialization vulnerability allowing unauthenticated attackers to read arbitrary server files by chaining io.FileIO and urllib.request.urlopen. Attackers can bypass RCE-focused blocklists to exfiltrate sensitive data like /etc/passwd to external servers.

Exploitation Scenario

An adversary publishes a poisoned model checkpoint to a public repository (e.g., a Hugging Face-style .pkl file with a plausible model card). When a victim's CI/CD pipeline runs picklescan on this artifact during an automated intake scan, the crafted pickle payload deserializes and executes a chain that opens /etc/passwd, ~/.aws/credentials, or a .env file via io.FileIO, then transmits the file contents to an attacker-controlled endpoint using urllib.request.urlopen. The payload deliberately avoids code execution patterns flagged by picklescan's RCE blocklist — it only reads and exfiltrates — so the scanner reports the file as clean while the exfiltration completes. The entire attack is fully automated and requires no interaction beyond the scheduled scan job triggering.

Weaknesses (CWE)

CWE-22 Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal') CWE-22 Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal')

CWE-22 — Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal'): The product uses external input to construct a pathname that is intended to identify a file or directory that is located underneath a restricted parent directory, but the product does not properly neutralize special elements within the pathname that can cause the pathname to resolve to a location that is outside of the restricted directory.

[Implementation] Assume all input is malicious. Use an "accept known good" input validation strategy, i.e., use a list of acceptable inputs that strictly conform to specifications. Reject any input that does not strictly conform to specifications, or transform it into something that does. When performing input validation, consider all potentially relevant properties, including length, type of input, the full range of acceptable values, missing or extra inputs, syntax, consistency across related fields, and conformance to business rules. As an example of business rule logic, "boat" may be syntactically valid because it only contains alphanumeric characters, but it is not valid if the input is only expected to contain colors such as "red" or "blue." Do not rely exclusively on looking for malicious or malformed inputs. This is likely to miss at least one undesirable input, especially if the code's environment changes. This can give attackers enough room to bypass the intended validation. However, denylis
[Architecture and Design] For any security checks that are performed on the client side, ensure that these checks are duplicated on the server side, in order to avoid CWE-602. Attackers can bypass the client-side checks by modifying values after the checks have been performed, or by changing the client to remove the client-side checks entirely. Then, these modified values would be submitted to the server.

Source: MITRE CWE corpus.