Atomicorp WAF Research Note: CVE-2026-47391

PraisonAI’s unauthenticated A2A official example can reach real LLM-driven eval() tool execution

CVE Details

• Already protected by existing Atomicorp WAF rules
• Protection status: Understanding Atomicorp WAF Protection Status
• CVE: CVE-2026-47391
• Name: PraisonAI’s unauthenticated A2A official example can reach real LLM-driven eval() tool execution
• Severity: Critical
• CVSS: 9.8 (v3.1) (Critical)
• CVSS version: 3.1
• CVSS source: GitHub_M CNA
• CVSS vector: CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H
  • Attack vector: Network
  • Attack complexity: Low
  • Privileges required: None
  • User interaction: None
  • Scope: Unchanged
  • Confidentiality impact: High
  • Integrity impact: High
  • Availability impact: High
• CWE: CWE-95 — Improper Neutralization of Directives in Dynamically Evaluated Code (‘Eval Injection’), CWE-306 — Missing Authentication for Critical Function
• Related CAPEC attack patterns (via MITRE CWE relationships): CAPEC-12 (opens in a new tab) , CAPEC-166 (opens in a new tab) , CAPEC-216 (opens in a new tab) , CAPEC-35 (opens in a new tab) , CAPEC-36 (opens in a new tab) , CAPEC-62 (opens in a new tab)
• Vendor: MervinPraison
• Vulnerable product: PraisonAI
• CVE published: 2026-07-21
• CVE last modified: 2026-07-23
• EPSS: 0.0117; 65.4th percentile (FIRST (opens in a new tab) )
• Latest Update: 2026-09-06
• Selected Rule IDs observed: 340014 , 340023 , 340029 , 344360 , 344361 , 344363 , 344364 , 344366 , 344370 , 393655
• Matching rules: 10; rules with blocking actions: 10
• Would-block Rule IDs: 340014 , 340023 , 340029 , 344360 , 344361 , 344363 , 344364 , 344366 , 344370 , 393655
• Request surfaces inspected: JSON request data, Request URI, Request argument names, Request arguments, Request cookies, SOAP request data, XML request data

Vulnerability Description

PraisonAI is a multi-agent teams system. Prior to version 4.6.40, PraisonAI’s first-party A2A server example exposes an unauthenticated A2A JSON-RPC endpoint and registers a calculate(expression) tool implemented with Python eval(). The example also binds to 0.0.0.0. A remote unauthenticated attacker can send message/send to /a2a; the request reaches agent.chat(), and a real LLM can invoke the registered calculate tool. In testing with gemini/gemini-2.5-flash-lite, this resulted in arbitrary Python execution in the server process, confirmed by creation of a marker file from an unauthenticated HTTP request. The issue affects deployments following the official A2A example or similar unauthenticated public A2A deployments with unsafe tools. The default unauthenticated A2A surface also exposes task history and task cancellation APIs, increasing confidentiality and integrity impact. Version 4.6.40 patches the issue.

Recommended Actions

1. Maintain the observed Atomicorp protections. Keep your Atomicorp commercial rules current and confirm that the selected observed rules (340014 , 340023 , 340029 , 344360 , 344361 , 344363 , 344364 , 344366 , 344370 , 393655 ) have not been locally disabled. Atomicorp commercial rules are enabled by default. If you use Atomic WAF, review the applicable policy and rule settings.

2. Apply the vendor’s remediation. WAF protection is defense in depth and does not replace upgrading or following the affected vendor’s guidance.

   • Affected versions: MervinPraison PraisonAI version < 4.6.40.
   • No specific CNA/vendor solution or workaround is included in the current CVE Record (opens in a new tab) . Review the vendor’s security advisories and supported-release guidance before making changes.

3. Validate after changes. Confirm rule updates are current, review WAF events for the affected application, and test application behavior after vendor upgrades, mitigations, or local policy changes.

Research Notes

Atomicorp WAF protections are designed around attack techniques and exploit behaviors, allowing existing protections to apply across multiple vulnerabilities without requiring a unique rule for every CVE.

Public Vulnerability References

• NVD (opens in a new tab)
• CVE Record (opens in a new tab)
• CVE authority reference: https://github.com/MervinPraison/PraisonAI/commit/e0fb8e7dd1ee6759c18ed07f436c21dbd9c20747 (opens in a new tab)
• CVE authority reference: https://github.com/MervinPraison/PraisonAI/pull/1793 (opens in a new tab)
• CVE authority reference: https://github.com/MervinPraison/PraisonAI/security/advisories/GHSA-vg22-4gmj-prxw (opens in a new tab)

Test or Research Source

• Source: Atomicorp AI Research Framework
• Tags: cve, derived, rce, waf-probe

Research Note Context

This research note documents exploit-technique analysis associated with this vulnerability and references Atomicorp WAF protections observed during engineering review, testing, or validation activities.

Atomicorp security controls are generally designed to identify and mitigate attack techniques and exploit behaviors rather than individual CVE identifiers. As a result, protections developed for one vulnerability may also apply to related attack variants and future vulnerabilities using similar techniques.

Learn More

Atomicorp develops commercially supported security technologies for web applications, servers, endpoints, workloads, and legacy enterprise platforms.

Organizations interested in protecting systems against exploit techniques associated with vulnerabilities such as this one can learn more about:

• Atomic ModSecurity Rules: commercially maintained WAF rules for ModSecurity deployments, including protections for web attacks, bots, scanners, malware, exploitation attempts, and emerging vulnerability classes.
• Atomic WAF: an enterprise web application firewall with Atomicorp rules, centralized management, reporting, authentication controls, and operational tooling.
• Atomic ModSecurity Integrator: a special licensing option for MSPs, MSSPs, OEMs, VARs, reverse proxy providers, and other integrators who want to incorporate Atomicorp ModSecurity rule technology into their own products or services.
• Atomic OSSEC: an enterprise OSSEC-based intrusion detection, compliance, and security monitoring platform.
• Atomic Protector: workload protection for Linux, web, cloud, data center, and hybrid environments.

For additional information, product evaluations, or technical discussions, visit https://www.atomicorp.com/ or https://modsecurity.io/ (opens in a new tab) .