/

CVE-2026-28448

Severity:

HIGH RISK

(7.3)

Access Control Bypass in OpenClaw Twitch Plugin Agent

CVE-2026-28448

Analyzed

Vulnerability type

Category:

Access Control Bypass

CWE:

CWE-285

Impact:

Unauthorized Agent Invocation and Potential Resource Exhaustion

Severity:

HIGH RISK

(7.3)

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

OpenClaw's Twitch plugin fails to enforce the allowFrom allowlist whenever allowedRoles is unset or empty, effectively disabling intended access restrictions.

In this misconfigured state, any remote Twitch user who can mention the bot in chat may trigger agent dispatch and invoke the agent pipeline, which can lead to unintended actions or resource exhaustion.

Affected context

This vulnerability appears in OpenClaw deployments that:

  • Run versions 2026.1.29 prior to 2026.2.1
  • Have the Twitch plugin installed, configured, and enabled
  • Rely on allowFrom plus allowedRoles for chat-based access control

The issue is exposed when allowedRoles is left unset, misconfigured, or explicitly set to an empty list. Under these conditions, the plugin does not properly apply the allowFrom allowlist, leaving bot mention-based agent dispatch exposed to any Twitch user interacting in the configured channel(s).

Exploit preconditions

  • Attack surface: Twitch chat interface where the OpenClaw bot is present and listening.
  • Required access: Ability to send messages in the Twitch channel(s) monitored by the OpenClaw bot (no special Twitch privileges required beyond normal chat participation).
  • Trigger point: Mentioning the bot in chat and issuing commands that would normally be restricted by allowFrom and role-based checks.
  • Why exploitation is possible: The authorization logic treats an unset or empty allowedRoles as a state where the allowFrom allowlist is not enforced, effectively short-circuiting access control and allowing any user mention to initiate agent dispatch.

Why this happens

The underlying issue is an authorization logic flaw consistent with CWE-285: Improper Authorization.

Instead of treating an unset or empty allowedRoles configuration as a hardened, deny-by-default state, the Twitch plugin behaves as if all users are permitted or as if the allowFrom restrictions do not apply.

This reflects a common bug pattern where guard conditions are only evaluated when certain configuration fields are populated, leaving an implicit "open" mode when those fields are empty. In secure designs, the absence of explicit permissions should default to no access, but here it results in unrestricted agent dispatch via public chat.

Potential impact

  • Unauthorized Twitch users may trigger agent pipelines that should be restricted to specific users or roles.
  • Attackers can cause repeated agent dispatches, leading to resource exhaustion (e.g., API quota consumption, CPU/memory load, or downstream service overuse).
  • Unintended automated actions could occur if agents perform external operations (such as posting content, calling APIs, or modifying data) based on chat commands.
  • Increased operational noise and log volume, potentially obscuring legitimate activity and complicating incident response.

General mitigation guidance

Effective mitigation focuses on enforcing strict, fail-closed authorization semantics and clarifying configuration behavior:

  • Design the authorization layer so that unset or empty allowedRoles results in denial of all non-admin or external requests, unless explicitly permitted by a clearly separate configuration flag.
  • Ensure that allowlists like allowFrom are always evaluated independently, and that their enforcement does not silently depend on other optional fields being populated.
  • Introduce clear configuration validation that warns or rejects deployments where access control settings are ambiguous or effectively open (for example, when allowedRoles is empty in production settings).
  • Apply principle of least privilege to agent actions, so that even when agent dispatch is triggered, the agents operate with constrained capabilities and resource limits.
  • Provide secure-by-default templates or presets where Twitch integration is restricted to known accounts or roles, requiring explicit opt-in to broaden access.

How teams detect issues like this

To detect and manage this class of issue, teams can consider:

  • Configuration audits: Automatically scan OpenClaw/Twitch plugin configuration files or environment variables for risky patterns, such as empty or missing allowedRoles in combination with enabled Twitch dispatch.
  • Static analysis and code review: Look for authorization checks that are conditional on configuration fields being non-empty. Flag patterns where the absence of configuration leads to permissive behavior instead of denial.
  • Runtime monitoring: Instrument logging for agent dispatch events including the originating Twitch user, channel, and applied authorization decision. Alert on dispatches originating from users not expected to have access.
  • Policy enforcement in CI/CD: Add automated checks that block deployments where access control parameters are misconfigured or left at permissive defaults, especially in production environments.
  • Behavioral anomaly detection: Track spikes in agent invocation volume or unusual command patterns from new or low-reputation accounts that may indicate automated abuse.

Sources

SourceLink
github.comgithub.com/openclaw/openclaw/commit/8c7901c984866a776eb59662dc9d8b028de4f0d0
github.comgithub.com/openclaw/openclaw/security/advisories/GHSA-33rq-m5x2-fvgf
vulncheck.comvulncheck.com/advisories/openclaw-authorization-bypass-in-twitch-plugin-allowfrom-access-control

Frequently asked

questions

Yes. Any Twitch user who can post messages in a channel monitored by the OpenClaw bot can potentially exploit this issue if the Twitch plugin is enabled and misconfigured (with allowedRoles unset or empty).

The attacker only needs standard Twitch chat access in the relevant channel. No local access to the OpenClaw host, admin panel, or infrastructure is required; interaction occurs entirely via Twitch chat.

Implement a deny-by-default authorization model: treat missing or empty allowedRoles as "no one is allowed" and always enforce allowFrom and other allowlists regardless of role configuration. Add explicit configuration validation to prevent ambiguous or open states.

Adopt a consistent access control design where:

  • Defaults are restrictive rather than permissive.
  • Authorization checks are centralized and independent of optional configuration fields.
  • Configurations are validated at startup and in CI to catch empty or unsafe settings.
  • Code reviews explicitly examine authorization edge cases, including missing or empty configuration values.
Find out what's already
exploitable in your codebase.