CVE-2026-61433: PraisonAI code injection

CISO Take

PraisonAI, an AI agent orchestration framework, fails to safely encode configuration values (deploy.api.host, agents_file) when it generates the Python source code for API deployment servers, letting an attacker smuggle arbitrary Python expressions into that generated code. The CVSS 7.8 score reflects full confidentiality, integrity, and availability impact once the injected code runs, which matters most for any team that builds agent deployment pipelines where config values can originate from a lower-trust source, such as a self-service agent builder, CI job, or multi-tenant platform. There is no EPSS score, no CISA KEV listing, no public exploit, and no Nuclei template for this CVE, and exploitation requires local access plus user interaction, so this is not an internet-mass-exploitation scenario today — but the blast radius is a full code-execution primitive on whatever host runs the generated server. Patch to PraisonAI 4.6.78 or later (fix commit 1620b49 referenced in the GHSA advisory); until then, treat deploy.api.host and agents_file as untrusted input, validate/allowlist them before deployment generation, and review generated server source before executing it.

Sources: NVD GitHub Advisory ATLAS vulncheck.com

What is the risk?

Moderate risk in the near term, high risk in specific deployment patterns. The CVSS 7.8 (AV:L/AC:L/PR:N/UI:R/C:H/I:H/A:H) means exploitation is technically easy once an attacker can influence the two config parameters, and no privileges are required — but the attack vector is local and requires a user to trigger the generated server, which caps mass exploitability. No EPSS percentile, KEV listing, public exploit, or scanner template exists, indicating this is not yet weaponized in the wild. Risk rises sharply for organizations that expose PraisonAI's deployment/config generation to less-trusted input paths (multi-tenant agent builders, self-service developer portals, automated CI/CD that consumes external config), where the "local" and "user interaction" barriers effectively evaporate.

How does the attack unfold?

Config injection

Attacker supplies a malicious value for deploy.api.host or agents_file in the PraisonAI deployment configuration.

AML.T0081

Unsafe code generation

PraisonAI generates Python source for the API server, embedding the unsanitized config value directly into executable code.

AML.T0050

Triggered execution

A user or admin runs the generated server, or the server handles an incoming request, causing the injected Python expression to execute.

AML.T0011

Impact

The attacker gains arbitrary code execution on the host, with full confidentiality, integrity, and availability compromise of the agent deployment.

Config injection

Attacker supplies a malicious value for deploy.api.host or agents_file in the PraisonAI deployment configuration.

AML.T0081

Unsafe code generation

PraisonAI generates Python source for the API server, embedding the unsanitized config value directly into executable code.

AML.T0050

Triggered execution

A user or admin runs the generated server, or the server handles an incoming request, causing the injected Python expression to execute.

AML.T0011

Impact

The attacker gains arbitrary code execution on the host, with full confidentiality, integrity, and availability compromise of the agent deployment.

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
PraisonAI	pip	—	No patch
1 dependents 69% patched ~13d to patch Full package profile →

Do you use PraisonAI? You're affected.

How severe is it?

CVSS 3.1

7.8 / 10

EPSS

0.2%

chance of exploitation in 30 days

Higher than 11% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

Exploit Available

Exploitation: MEDIUM

Sophistication

Moderate

Exploitation Confidence

medium

○ 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 High

A High

What should I do?

1 step

Upgrade to PraisonAI 4.6.78 or later (fix referenced in commit 1620b49 on GHSA-79fv-7hq9-w7xg). Until patched: never accept deploy.api.host or agents_file values from untrusted or external sources without strict validation/allowlisting; treat any generated Python server source as untrusted output and review it before execution; restrict who can invoke PraisonAI's deployment/codegen feature (principle of least privilege); and add detection for anomalous process spawns or unexpected outbound connections originating from PraisonAI-generated API server processes.

What does CISA's SSVC say?

Decision Track

Exploitation none

Automatable No

Technical Impact total

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

How is it classified?

Code Execution Framework Agent API AML.T0011 - User Execution AML.T0050 - Command and Scripting Interpreter 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

NIST AI RMF

MANAGE-2.3 - Third-party AI system risks are managed

OWASP LLM Top 10

LLM02 - Insecure Output Handling

Frequently Asked Questions

What is CVE-2026-61433?

PraisonAI, an AI agent orchestration framework, fails to safely encode configuration values (deploy.api.host, agents_file) when it generates the Python source code for API deployment servers, letting an attacker smuggle arbitrary Python expressions into that generated code. The CVSS 7.8 score reflects full confidentiality, integrity, and availability impact once the injected code runs, which matters most for any team that builds agent deployment pipelines where config values can originate from a lower-trust source, such as a self-service agent builder, CI job, or multi-tenant platform. There is no EPSS score, no CISA KEV listing, no public exploit, and no Nuclei template for this CVE, and exploitation requires local access plus user interaction, so this is not an internet-mass-exploitation scenario today — but the blast radius is a full code-execution primitive on whatever host runs the generated server. Patch to PraisonAI 4.6.78 or later (fix commit 1620b49 referenced in the GHSA advisory); until then, treat deploy.api.host and agents_file as untrusted input, validate/allowlist them before deployment generation, and review generated server source before executing it.

Is CVE-2026-61433 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2026-61433, increasing the risk of exploitation.

How to fix CVE-2026-61433?

Upgrade to PraisonAI 4.6.78 or later (fix referenced in commit 1620b49 on GHSA-79fv-7hq9-w7xg). Until patched: never accept deploy.api.host or agents_file values from untrusted or external sources without strict validation/allowlisting; treat any generated Python server source as untrusted output and review it before execution; restrict who can invoke PraisonAI's deployment/codegen feature (principle of least privilege); and add detection for anomalous process spawns or unexpected outbound connections originating from PraisonAI-generated API server processes.

What systems are affected by CVE-2026-61433?

This vulnerability affects the following AI/ML architecture patterns: agent frameworks, agent deployment pipelines, API server generation.

What is the CVSS score for CVE-2026-61433?

CVE-2026-61433 has a CVSS v3.1 base score of 7.8 (HIGH). The EPSS exploitation probability is 0.21%.

What is the AI security impact?

Affected AI Architectures

agent frameworksagent deployment pipelinesAPI server generation

MITRE ATLAS Techniques

AML.T0011 User Execution

AML.T0050 Command and Scripting Interpreter

AML.T0081 Modify AI Agent Configuration

Compliance Controls Affected

EU AI Act: Article 15

NIST AI RMF: MANAGE-2.3

OWASP LLM Top 10: LLM02

What are the technical details?

Original Advisory

PraisonAI before 4.6.78 fails to safely encode deployment configuration values when generating Python source code for API servers. Attackers can inject arbitrary Python expressions through the deploy.api.host and agents_file configuration parameters that execute when the generated server starts or handles requests.

Exploitation Scenario

A developer or tenant on a platform that exposes PraisonAI's deployment configuration submits a crafted value for deploy.api.host — or points agents_file at a config containing an embedded Python expression. PraisonAI's code generator writes this value directly into the Python source of the generated API server without sanitization. When an administrator later runs that generated server (satisfying the user-interaction requirement) or when the server processes an inbound request that reaches the tainted code path, the injected expression executes with the privileges of the server process, giving the attacker arbitrary code execution and a foothold to access agent credentials, connected tools, or the underlying host.

Weaknesses (CWE)

CWE-94 Improper Control of Generation of Code ('Code Injection') Primary CWE-94 Improper Control of Generation of Code ('Code Injection')

CWE-94 — Improper Control of Generation of Code ('Code Injection'): The product constructs all or part of a code segment using externally-influenced input from an upstream component, but it does not neutralize or incorrectly neutralizes special elements that could modify the syntax or behavior of the intended code segment.

[Architecture and Design] Refactor your program so that you do not have to dynamically generate code.
[Architecture and Design] Run your code in a "jail" or similar sandbox environment that enforces strict boundaries between the process and the operating system. This may effectively restrict which code can be executed by your product. Examples include the Unix chroot jail and AppArmor. In general, managed code may provide some protection. This may not be a feasible solution, and it only limits the impact to the operating system; the rest of your application may still be subject to compromise. Be careful to avoid CWE-243 and other weaknesses related to jails.

Source: MITRE CWE corpus.