/

CVE-2026-57140

Severity:

CRITICAL RISK

(9.4)

Missing Authentication (CWE-306) in PraisonAI AgentOS

CVE-2026-57140

Deferred

Vulnerability type

Category:

Missing Authentication

CWE:

CWE-306

Impact:

Unauthorized Agent Invocation and Sensitive Resource Access

Severity:

CRITICAL RISK

(9.4)

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

PraisonAI AgentOS exposes agent discovery and chat endpoints without authentication middleware, allowing reachable remote callers to inspect configured agents and invoke them. Depending on the agent configuration, this may provide access to tools, memory, external APIs, credentials, or workflow state.

Affected context

The vulnerability affects PraisonAI deployments using AgentOS versions from 1.6.0 through 1.7.2 where the service binds to the default 0.0.0.0 address and registers agent and chat routes without authentication middleware. Risk is greatest when the service is reachable from an untrusted network and configured agents have access to privileged tools, secrets, external APIs, or persistent workflow data.

Exploit preconditions

  • Attack surface: Network-accessible AgentOS HTTP service
  • Required access: Network connectivity to the exposed service; valid application authentication is not required at the affected routes
  • Trigger point: Requests to agent listing or chat functionality
  • Why exploitation is possible: The routes are registered without an authentication control, allowing remote callers to enumerate agents and submit requests for agent execution

Why this happens

The underlying issue is missing authentication at sensitive HTTP endpoints, mapping to CWE-306: Missing Authentication for Critical Function. AgentOS exposes functionality that can disclose internal agent metadata and initiate potentially privileged actions, but the route handling path does not require the caller to establish an authenticated identity before access is granted. Binding the service to a broadly reachable address increases exposure but does not replace endpoint-level authorization.

Potential impact

  • Unauthorized callers may enumerate agent names, roles, and instruction prefixes
  • Attackers could invoke selected agents without authorization
  • Agent tools may be used to access external services or connected systems
  • Sensitive credentials, memory, or workflow state could be exposed
  • Agent behavior or downstream data may be altered
  • The service could be used as a stepping stone for broader compromise

General mitigation guidance

Mitigation generally focuses on enforcing authentication and authorization before any agent discovery or invocation occurs.

Common defensive patterns include:

  • Require authentication middleware on every sensitive route by default
  • Apply explicit authorization checks to agent listing, selection, tool use, and workflow operations
  • Prefer deny-by-default route policies and secure service binding defaults
  • Restrict network exposure through private interfaces, ingress controls, and allowlists
  • Keep credentials and high-impact tools outside agents that do not require them
  • Run agent processes with least-privilege identities and narrowly scoped permissions
  • Avoid returning sensitive agent metadata to unauthenticated or unauthorized callers
  • Separate administrative, discovery, and execution capabilities into distinct trust boundaries

How teams detect issues like this

Organizations can identify similar issues through:

  • Static analysis and route audits for handlers lacking authentication or authorization middleware
  • Code review checks for newly registered API endpoints and default network bindings
  • Integration tests that verify unauthenticated requests are rejected for sensitive operations
  • API gateway and policy rules requiring identity checks before agent discovery or execution
  • Runtime monitoring for anonymous requests, unusual agent enumeration, and unexpected tool invocation
  • Configuration reviews that inventory exposed listeners and connected agent capabilities

Automated security review tools may help flag unauthenticated route registration, but endpoint behavior should also be validated with authorization-focused tests.

Sources

SourceLink
github.comgithub.com/MervinPraison/PraisonAI/commit/709a0387853709a5e2f2c34680fa64cada252c27
github.comgithub.com/MervinPraison/PraisonAI/releases/tag/v4.6.60
github.comgithub.com/MervinPraison/PraisonAI/security/advisories/GHSA-9752-mhqh-h34f

Frequently asked

questions

Yes, when the AgentOS service is reachable by an attacker. Exploitation does not require local access if the affected routes are exposed over the network.

The affected endpoints do not enforce authentication middleware, so a caller who can reach the service may be able to access them without application credentials.

Require authentication and authorization before agent discovery, chat requests, tool execution, and access to connected resources. Network isolation should provide defense in depth rather than replace endpoint controls.

Use deny-by-default API policies, test every sensitive route for anonymous access, minimize agent privileges, and continuously monitor exposed service interfaces and agent capabilities.

Find out what's already
exploitable in your codebase.