/

CVE-2026-72776

Severity:

CRITICAL RISK

(9.8)

Unauthenticated RCE (CWE-306) in AgenticSeek API Endpoint

CVE-2026-72776

Deferred

Vulnerability type

Category:

Missing Authentication for Critical Function

CWE:

CWE-306

Impact:

Unauthenticated Remote Code Execution

Severity:

CRITICAL RISK

(9.8)

CVSS version:

3.1

Catch this CVE while code is being written

Detect vulnerable code directly in pull requests, so CVEs are fixed before merge.

Summary

AgenticSeek exposes an unprotected query endpoint that allows network-adjacent users to submit requests without authentication. Crafted queries may cause the autonomous agent to execute attacker-controlled shell commands, resulting in arbitrary code execution on the host.

Affected context

The vulnerability occurs in deployments where AgenticSeek exposes the POST /query API on a network-accessible interface, including configurations bound to 0.0.0.0:7777. Risk increases when the endpoint lacks authentication, permits broad cross-origin access, and routes agent-generated commands to BashInterpreter with shell execution enabled.

Exploit preconditions

  • Attack surface: Network-accessible HTTP POST /query endpoint
  • Required access: Network adjacency or reachability; no authentication is required
  • Trigger point: Processing of a query by the autonomous agent and its command-execution workflow
  • Why exploitation is possible: The endpoint accepts unauthenticated input, while downstream command generation and execution can invoke shell commands with insufficiently reliable safety controls

Why this happens

The primary weakness is missing authentication for a security-sensitive function, corresponding to CWE-306. A publicly reachable API performs privileged agent operations without verifying the caller's identity or authorization. The risk is compounded by an insecure command-execution design in which generated commands are passed to a shell through process execution, while a partial blocklist is treated as a safety boundary. This combination allows untrusted requests to reach host-level execution logic.

Potential impact

  • An attacker may execute arbitrary commands on the AgenticSeek host
  • The application environment could be fully compromised
  • Credentials, files, and environment secrets may be exposed
  • Attackers could modify data, install persistence, or disrupt services
  • The compromised host could be used for lateral movement or further attacks

General mitigation guidance

Mitigation generally focuses on enforcing authentication and authorization before any agent or command-execution operation is reached.

Common defensive patterns include:

  • Require strong authentication and explicit authorization for the query endpoint
  • Bind administrative or execution-capable services to restricted interfaces by default
  • Avoid shell interpretation and prefer fixed executable paths with structured argument arrays
  • Replace generated command execution with narrowly scoped, allowlisted capabilities
  • Treat blocklists as insufficient and reduce trust in all user-controlled or agent-generated input
  • Apply least-privilege identities, filesystem permissions, and process isolation to the agent
  • Restrict cross-origin access to approved origins and enforce network-level access controls
  • Add confirmation or policy gates before sensitive operations are executed

How teams detect issues like this

High-level detection approaches include:

  • Static analysis for unauthenticated routes that invoke subprocesses, shell interpreters, or agent tools
  • Code review checks for shell=True, dynamic command construction, and blocklist-only validation
  • API inventory reviews that identify execution-capable endpoints exposed on wildcard interfaces
  • Runtime monitoring for unusual requests to /query, unexpected child processes, and shell activity from the agent service
  • Policy enforcement requiring authentication, authorization, and restricted network exposure for privileged APIs
  • Automated security review tools can help flag missing access controls and unsafe process-execution patterns

Sources

SourceLink
github.comgithub.com/Fosowl/agenticSeek
github.comgithub.com/Fosowl/agenticSeek/issues/520
github.comgithub.com/Fosowl/agenticSeek/pull/534/changes/f1eb2cfc721f8a21dd16a8b048a9ca89f3259f6f
vulncheck.comvulncheck.com/advisories/agenticseek-unauthenticated-rce-via-query-api-endpoint

Frequently asked

questions

Yes. A network-adjacent attacker may be able to reach the exposed POST /query endpoint and submit requests without authenticating.

No account is required when the endpoint is deployed without authentication. Network reachability and the ability to send HTTP requests are the key prerequisites.

Require authentication and authorization before processing queries, and remove general-purpose shell execution in favor of narrowly scoped, structured operations.

Secure execution-capable APIs by default, avoid shell interpretation of dynamic input, enforce least privilege and network restrictions, and continuously test access-control and command-execution policies.

Find out what's already
exploitable in your codebase.