CVE-2026-61431: PraisonAI path traversal leaks

CISO Take

PraisonAI's ContextGatherer builds an AI agent's working context from .praisoncontext and .praisoninclude files without validating the include paths, so an attacker who controls one of those files can point it at absolute paths or use ../ sequences to pull arbitrary host files — SSH keys, .env secrets, credentials — into the generated context bundle. There's no EPSS score, no CISA KEV listing, no public exploit or Nuclei template, and exploitation requires local access plus a victim action (running PraisonAI against an attacker-supplied repo or config), which keeps near-term mass-exploitation risk low. The real exposure is anyone who runs PraisonAI agents against third-party or untrusted repositories — CI pipelines, contributor onboarding, or shared agent configs are the likely vector, and the confidentiality impact is high (CVSS 5.5, C:H/I:N/A:N) because file contents can flow into whatever the agent does with its context, including onward LLM API calls. Patch to PraisonAI 4.6.78 or later, and until then treat any .praisoncontext/.praisoninclude file from an external repo as untrusted input requiring review before running the agent.

Sources: NVD GitHub Advisory ATLAS vulncheck.com

What is the risk?

Medium severity (CVSS 5.5) driven entirely by confidentiality impact — no integrity or availability effect. Attack vector is local and requires user interaction, which meaningfully limits remote/drive-by exploitation, but attack complexity is low and no privileges are required once a victim runs PraisonAI against a malicious repo or config. No EPSS percentile, no KEV listing, no public PoC or scanner template exist yet, so real-world exploitation likelihood is currently low — but the bug class (unsanitized include-path resolution) is trivial to weaponize once someone writes a PoC, and the payoff (SSH keys, .env secrets, cloud credentials) is high value for supply-chain-style attacks against developers and CI runners.

How does the attack unfold?

Malicious config distribution

Attacker publishes or contributes a repo containing a crafted .praisoncontext or .praisoninclude file with absolute paths or '../' traversal sequences pointing at sensitive host files.

AML.T0081

Victim execution

A developer or CI pipeline clones the repo and runs a PraisonAI agent against it, invoking ContextGatherer on the untrusted include files.

AML.T0011

Path traversal read

ContextGatherer resolves the unsanitized include paths and reads files outside the intended workspace, such as SSH keys or .env secrets.

AML.T0037

Context exposure

The exfiltrated file contents are bundled into the generated context, potentially surfaced in agent output or forwarded to downstream LLM API calls, giving the attacker access to the leaked data.

AML.T0025

Malicious config distribution

Attacker publishes or contributes a repo containing a crafted .praisoncontext or .praisoninclude file with absolute paths or '../' traversal sequences pointing at sensitive host files.

AML.T0081

Victim execution

A developer or CI pipeline clones the repo and runs a PraisonAI agent against it, invoking ContextGatherer on the untrusted include files.

AML.T0011

Path traversal read

ContextGatherer resolves the unsanitized include paths and reads files outside the intended workspace, such as SSH keys or .env secrets.

AML.T0037

Context exposure

The exfiltrated file contents are bundled into the generated context, potentially surfaced in agent output or forwarded to downstream LLM API calls, giving the attacker access to the leaked data.

AML.T0025

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
PraisonAI	pip	—	No patch
1 dependents 69% patched ~12d to patch Full package profile →

Do you use PraisonAI? You're affected.

How severe is it?

CVSS 3.1

5.5 / 10

EPSS

0.4%

chance of exploitation in 30 days

Higher than 28% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

Exploit Available

Exploitation: MEDIUM

Sophistication

Moderate

Exploitation Confidence

medium

○ CISA SSVC: Public PoC

○ Public PoC indexed (trickest/cve)

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 Local

AC Low

PR None

UI Required

S Unchanged

C High

I None

A None

What should I do?

1 step

Upgrade to PraisonAI 4.6.78 or later, which fixes the path validation in ContextGatherer. Until patched, treat .praisoncontext and .praisoninclude files from any repo you didn't author as untrusted — review them before running PraisonAI, and avoid running the agent against unvetted third-party repos or CI jobs that check out arbitrary branches/PRs. Where possible, run PraisonAI in a sandboxed/containerized workspace with filesystem access scoped to the project directory only, so even an unpatched traversal can't reach host secrets. For detection, monitor for ContextGatherer reads outside the working directory (e.g., paths resolving to home directory, /etc, or parent directories) and flag .praisoncontext/.praisoninclude files containing absolute paths or '../' sequences in code review or pre-commit hooks.

What does CISA's SSVC say?

Decision Track*

Exploitation poc

Automatable No

Technical Impact partial

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

How is it classified?

Data Leakage Supply Chain Agent Framework AML.T0011 - User Execution AML.T0025 - Exfiltration via Cyber Means AML.T0037 - Data from Local System AML.T0081 - Modify AI Agent Configuration

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 15 - Accuracy, robustness and cybersecurity

ISO 42001

A.6.2.6 - AI system security controls

NIST AI RMF

MEASURE 2.7 - AI system security and resilience testing

OWASP LLM Top 10

LLM02:2025 - Sensitive Information Disclosure LLM08:2025 - Excessive Agency

Frequently Asked Questions

What is CVE-2026-61431?

PraisonAI's ContextGatherer builds an AI agent's working context from .praisoncontext and .praisoninclude files without validating the include paths, so an attacker who controls one of those files can point it at absolute paths or use ../ sequences to pull arbitrary host files — SSH keys, .env secrets, credentials — into the generated context bundle. There's no EPSS score, no CISA KEV listing, no public exploit or Nuclei template, and exploitation requires local access plus a victim action (running PraisonAI against an attacker-supplied repo or config), which keeps near-term mass-exploitation risk low. The real exposure is anyone who runs PraisonAI agents against third-party or untrusted repositories — CI pipelines, contributor onboarding, or shared agent configs are the likely vector, and the confidentiality impact is high (CVSS 5.5, C:H/I:N/A:N) because file contents can flow into whatever the agent does with its context, including onward LLM API calls. Patch to PraisonAI 4.6.78 or later, and until then treat any .praisoncontext/.praisoninclude file from an external repo as untrusted input requiring review before running the agent.

Is CVE-2026-61431 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2026-61431, increasing the risk of exploitation.

How to fix CVE-2026-61431?

Upgrade to PraisonAI 4.6.78 or later, which fixes the path validation in ContextGatherer. Until patched, treat .praisoncontext and .praisoninclude files from any repo you didn't author as untrusted — review them before running PraisonAI, and avoid running the agent against unvetted third-party repos or CI jobs that check out arbitrary branches/PRs. Where possible, run PraisonAI in a sandboxed/containerized workspace with filesystem access scoped to the project directory only, so even an unpatched traversal can't reach host secrets. For detection, monitor for ContextGatherer reads outside the working directory (e.g., paths resolving to home directory, /etc, or parent directories) and flag .praisoncontext/.praisoninclude files containing absolute paths or '../' sequences in code review or pre-commit hooks.

What systems are affected by CVE-2026-61431?

This vulnerability affects the following AI/ML architecture patterns: agent frameworks, AI agent configuration, local context/RAG-style ingestion pipelines.

What is the CVSS score for CVE-2026-61431?

CVE-2026-61431 has a CVSS v3.1 base score of 5.5 (MEDIUM). The EPSS exploitation probability is 0.35%.

What is the AI security impact?

Affected AI Architectures

agent frameworksAI agent configurationlocal context/RAG-style ingestion pipelines

MITRE ATLAS Techniques

AML.T0011 User Execution

AML.T0025 Exfiltration via Cyber Means

AML.T0037 Data from Local System

AML.T0081 Modify AI Agent Configuration

Compliance Controls Affected

EU AI Act: Article 15

ISO 42001: A.6.2.6

NIST AI RMF: MEASURE 2.7

OWASP LLM Top 10: LLM02:2025, LLM08:2025

What are the technical details?

Original Advisory

PraisonAI before 4.6.78 contains a path traversal vulnerability in ContextGatherer that fails to validate include paths in .praisoncontext and .praisoninclude files. Attackers can supply absolute paths or parent directory traversal sequences to read arbitrary files outside the workspace and include their contents in the generated context bundle.

Exploitation Scenario

An attacker publishes or contributes to a repository that includes a crafted .praisoncontext (or .praisoninclude) file referencing paths like /home/victim/.ssh/id_rsa or ../../../.env using absolute paths or parent-directory traversal. A developer or CI job clones the repo and runs a PraisonAI agent against it as part of normal workflow; ContextGatherer resolves the unsanitized include paths and reads the referenced files, bundling their contents into the generated context. That context is then available to whatever the agent does next — surfaced in agent output, logged, or forwarded to an LLM API call — giving the attacker a path to exfiltrate SSH keys, API tokens, or other credentials without ever needing remote access to the victim's machine.

Weaknesses (CWE)

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