CVE-2026-31991: OpenClaw auth bypass — LOW

CISO Take

OpenClaw prior to 2026.2.26 allows an attacker who obtains a single direct-message pairing approval to bypass Signal group allowlist controls entirely, leaking a logical trust boundary that should separate DM-level identity from group-level authorization. Although CVSS scores this at 3.7 (low), the CVE ranks in the top 95th EPSS percentile for exploitation likelihood — and OpenClaw's skills ecosystem was already documented abusing unauthorized access vectors to distribute credential-stealing malware at scale, with 17% of surveyed skills classified as malicious (AIID #1368). In agentic deployments where group channels coordinate multi-agent workflows or contain operational data, this boundary flaw could expose task instructions, API keys, or sensitive business intelligence to an unauthorized party. Patch to 2026.2.26 immediately using either referenced commit; audit group membership against expected allowlists; until patched, restrict DM pairing approvals to explicitly vetted identities.

Sources: NVD EPSS GitHub Advisory VulnCheck ATLAS

What is the risk?

Moderate risk despite the low CVSS 3.7 rating. The top-95th-percentile EPSS score and documented ecosystem abuse in AIID #1368 elevate practical exploitation likelihood well above what CVSS alone implies. The attack requires low privileges, network access, and user interaction — the victim must approve a DM pairing request — creating a meaningful but not prohibitive barrier. Blast radius is limited to 4 known downstream dependents. In agentic orchestration deployments where OpenClaw group channels route sensitive tasks or carry credentials, the confidentiality and integrity impact is materially understated by the CVSS score. No public exploit or active exploitation confirmed at time of publication.

How does the attack unfold?

Initial Access via DM Pairing

Attacker initiates and obtains a DM pairing approval from a legitimate OpenClaw user through social engineering or exploitation of a low-trust relationship.

AML.T0012

Authorization Boundary Exploitation

Attacker exploits the CWE-863 flaw in OpenClaw's Signal group allowlist logic, presenting the DM pairing-store approval as valid group membership credentials to bypass the allowlist check.

AML.T0091

Unauthorized Group Access

Attacker gains read and write access to restricted group channels intended only for allowlisted members, bypassing the intended access control boundary.

AML.T0107

Data Collection and Agent Channel Abuse

Attacker harvests sensitive agent task assignments, operational data, or credentials from group channels, or injects malicious instructions into multi-agent coordination streams.

AML.T0085

Initial Access via DM Pairing

Attacker initiates and obtains a DM pairing approval from a legitimate OpenClaw user through social engineering or exploitation of a low-trust relationship.

AML.T0012

Authorization Boundary Exploitation

Attacker exploits the CWE-863 flaw in OpenClaw's Signal group allowlist logic, presenting the DM pairing-store approval as valid group membership credentials to bypass the allowlist check.

AML.T0091

Unauthorized Group Access

Attacker gains read and write access to restricted group channels intended only for allowlisted members, bypassing the intended access control boundary.

AML.T0107

Data Collection and Agent Channel Abuse

Attacker harvests sensitive agent task assignments, operational data, or credentials from group channels, or injects malicious instructions into multi-agent coordination streams.

AML.T0085

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
OpenClaw	pip	—	No patch
4 dependents 37% patched ~3d to patch Full package profile →
OpenClaw	pip	—	No patch
4 dependents 37% patched ~3d to patch Full package profile →

How severe is it?

CVSS 3.1

3.7 / 10

EPSS

0.2%

chance of exploitation in 30 days

Higher than 5% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

No known exploitation

Sophistication

Moderate

What is the attack surface?

AV Network

AC High

PR Low

UI Required

S Unchanged

C Low

I Low

A None

What should I do?

5 steps

Upgrade OpenClaw to 2026.2.26 or later (patch commits 8bdda7a651c21e98faccdbbd73081e79cffe8be0 and 64de4b6d6ae81e269ceb4ca16f53cda99ced967a).
Audit all group allowlist configurations to identify and remove any unexpected members that may have gained access prior to patching.
Review DM pairing-store entries and revoke any pairings from unrecognized or low-trust identities.
Monitor group access logs for anomalous sender identities, particularly any that appear to have joined via DM pairing rather than explicit group allowlist entry.
Until patched, implement compensating controls by restricting DM pairing approvals to a vetted allowlist of identities.

What does CISA's SSVC say?

Decision Track

Exploitation none

Automatable No

Technical Impact partial

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

How is it classified?

Auth Bypass Agent AML.T0012 - Valid Accounts AML.T0091 - Use Alternate Authentication Material AML.T0107 - Exploitation for Defense Evasion

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 9 - Risk management system

ISO 42001

A.6.2 - Security of AI systems

NIST AI RMF

GOVERN 6.1 - Policies and procedures for trustworthy AI

OWASP LLM Top 10

LLM02 - Sensitive Information Disclosure

Frequently Asked Questions

What is CVE-2026-31991?

OpenClaw prior to 2026.2.26 allows an attacker who obtains a single direct-message pairing approval to bypass Signal group allowlist controls entirely, leaking a logical trust boundary that should separate DM-level identity from group-level authorization. Although CVSS scores this at 3.7 (low), the CVE ranks in the top 95th EPSS percentile for exploitation likelihood — and OpenClaw's skills ecosystem was already documented abusing unauthorized access vectors to distribute credential-stealing malware at scale, with 17% of surveyed skills classified as malicious (AIID #1368). In agentic deployments where group channels coordinate multi-agent workflows or contain operational data, this boundary flaw could expose task instructions, API keys, or sensitive business intelligence to an unauthorized party. Patch to 2026.2.26 immediately using either referenced commit; audit group membership against expected allowlists; until patched, restrict DM pairing approvals to explicitly vetted identities.

Is CVE-2026-31991 actively exploited?

No confirmed active exploitation of CVE-2026-31991 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-31991?

1. Upgrade OpenClaw to 2026.2.26 or later (patch commits 8bdda7a651c21e98faccdbbd73081e79cffe8be0 and 64de4b6d6ae81e269ceb4ca16f53cda99ced967a). 2. Audit all group allowlist configurations to identify and remove any unexpected members that may have gained access prior to patching. 3. Review DM pairing-store entries and revoke any pairings from unrecognized or low-trust identities. 4. Monitor group access logs for anomalous sender identities, particularly any that appear to have joined via DM pairing rather than explicit group allowlist entry. 5. Until patched, implement compensating controls by restricting DM pairing approvals to a vetted allowlist of identities.

What systems are affected by CVE-2026-31991?

This vulnerability affects the following AI/ML architecture patterns: agent frameworks, multi-agent orchestration, communication-integrated agents.

What is the CVSS score for CVE-2026-31991?

CVE-2026-31991 has a CVSS v3.1 base score of 3.7 (LOW). The EPSS exploitation probability is 0.15%.

What is the AI security impact?

Affected AI Architectures

agent frameworksmulti-agent orchestrationcommunication-integrated agents

MITRE ATLAS Techniques

AML.T0012 Valid Accounts

AML.T0091 Use Alternate Authentication Material

AML.T0107 Exploitation for Defense Evasion

Compliance Controls Affected

EU AI Act: Article 9

ISO 42001: A.6.2

NIST AI RMF: GOVERN 6.1

OWASP LLM Top 10: LLM02

What are the technical details?

Original Advisory

OpenClaw versions prior to 2026.2.26 contain an authorization bypass vulnerability where Signal group allowlist policy incorrectly accepts sender identities from DM pairing-store approvals. Attackers can exploit this boundary weakness by obtaining DM pairing approval to bypass group allowlist checks and gain unauthorized group access.

Exploitation Scenario

An adversary targeting an organization that uses OpenClaw to coordinate AI agents via group channels initiates a DM pairing request with a legitimate OpenClaw participant — for example, a junior employee or a less security-aware agent operator. Once the target approves the DM pairing, a common and low-friction action, the attacker's identity is recorded in OpenClaw's DM pairing store. The attacker then exploits the allowlist boundary flaw (CWE-863: Incorrect Authorization) to present this DM-derived pairing approval as valid group membership, bypassing the group allowlist check. The attacker gains access to the group channel, where they can read agent task assignments, tool call results, or API keys shared in operational messages — or inject malicious instructions into the agent coordination stream, a scenario directly analogous to the credential-stealing abuse documented in AIID #1368.

Weaknesses (CWE)

CWE-863 Incorrect Authorization

CWE-863 — Incorrect Authorization: The product performs an authorization check when an actor attempts to access a resource or perform an action, but it does not correctly perform the check.

[Architecture and Design] Divide the product into anonymous, normal, privileged, and administrative areas. Reduce the attack surface by carefully mapping roles with data and functionality. Use role-based access control (RBAC) [REF-229] to enforce the roles at the appropriate boundaries. Note that this approach may not protect against horizontal authorization, i.e., it will not protect a user from attacking others with the same role.
[Architecture and Design] Ensure that access control checks are performed related to the business logic. These checks may be different than the access control checks that are applied to more generic resources such as files, connections, processes, memory, and database records. For example, a database may restrict access for medical records to a specific database user, but each record might only be intended to be accessible to the patient and the patient's doctor [REF-7].

Source: MITRE CWE corpus.