/

CVE-2026-123023

Severity:

HIGH RISK

(7.3)

Incorrect Authorization (CWE-863) in AgentScope Bash Tool

CVE-2026-123023

Deferred

Vulnerability type

Category:

Incorrect Authorization

CWE:

CWE-285, CWE-863

Impact:

Unauthorized execution of Bash Tool actions

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

AgentScope's Bash Tool has an authorization-checking flaw in its check_permissions function that can allow actions to proceed when they should be denied. Remote exploitation may be possible in deployments where an attacker can reach and invoke the tool; the exact access requirements depend on the deployment.

AgentScope's Bash Tool has an authorization-checking flaw in its check_permissions function that can allow actions to proceed when they should be denied. Remote exploitation may be possible in deployments where an attacker can reach and invoke the tool; the exact access requirements depend on the deployment.

Affected context

The issue affects the Bash Tool in AgentScope, specifically authorization checks performed by check_permissions in src/agentscope/tool/_builtin/_bash.py. Risk depends on whether the tool is available to remote users and how its permissions and access controls are configured.

The issue affects the Bash Tool in AgentScope, specifically authorization checks performed by check_permissions in src/agentscope/tool/_builtin/_bash.py. Risk depends on whether the tool is available to remote users and how its permissions and access controls are configured.

Exploit preconditions

  • Attack surface: An AgentScope deployment that exposes or makes the Bash Tool reachable
  • Required access: The attacker must be able to reach the relevant tool interface; the description does not specify whether authentication is required
  • Trigger point: An invocation that reaches the Bash Tool's permission-checking logic
  • Why exploitation is possible: Incorrect authorization can allow an action that should have been denied
  • Attack surface: An AgentScope deployment that exposes or makes the Bash Tool reachable
  • Required access: The attacker must be able to reach the relevant tool interface; the description does not specify whether authentication is required
  • Trigger point: An invocation that reaches the Bash Tool's permission-checking logic
  • Why exploitation is possible: Incorrect authorization can allow an action that should have been denied

Why this happens

The flaw is in authorization logic: the permission check does not correctly determine whether an action is allowed. This maps to CWE-863: Incorrect Authorization, a weakness in which access-control decisions may not match the intended policy. More broadly, it is an authorization enforcement failure, rather than a problem with the Bash command parser itself.

The flaw is in authorization logic: the permission check does not correctly determine whether an action is allowed. This maps to CWE-863: Incorrect Authorization, a weakness in which access-control decisions may not match the intended policy. More broadly, it is an authorization enforcement failure, rather than a problem with the Bash command parser itself.

Potential impact

  • An attacker may perform Bash Tool actions they are not authorized to perform
  • Systems or data accessible to the tool could be exposed or modified
  • The issue could provide a path to further compromise, depending on the tool's privileges and deployment boundaries
  • An attacker may perform Bash Tool actions they are not authorized to perform
  • Systems or data accessible to the tool could be exposed or modified
  • The issue could provide a path to further compromise, depending on the tool's privileges and deployment boundaries

General mitigation guidance

Mitigation generally focuses on making authorization decisions explicit, consistent, and fail-closed.

Common defensive patterns include:

  • Defining and enforcing a clear allowlist of permitted actions
  • Checking authorization at the point where each sensitive action is performed
  • Denying requests when identity, permission, or policy evaluation is missing or ambiguous
  • Limiting the Bash Tool's privileges and access to files, processes, and other resources
  • Restricting who can reach or invoke the tool and reducing trust in caller-supplied data

Mitigation generally focuses on making authorization decisions explicit, consistent, and fail-closed.

Common defensive patterns include:

  • Defining and enforcing a clear allowlist of permitted actions
  • Checking authorization at the point where each sensitive action is performed
  • Denying requests when identity, permission, or policy evaluation is missing or ambiguous
  • Limiting the Bash Tool's privileges and access to files, processes, and other resources
  • Restricting who can reach or invoke the tool and reducing trust in caller-supplied data

How teams detect issues like this

Look for authorization checks whose results can be bypassed, inconsistently applied, or fail open. Code reviews and static analysis can examine permission-checking logic, sensitive Bash Tool operations, and paths where access decisions are skipped or performed too late. Runtime policies can also restrict tool invocation to approved identities and actions, with logging and alerting for denied or unusual requests.

Look for authorization checks whose results can be bypassed, inconsistently applied, or fail open. Code reviews and static analysis can examine permission-checking logic, sensitive Bash Tool operations, and paths where access decisions are skipped or performed too late. Runtime policies can also restrict tool invocation to approved identities and actions, with logging and alerting for denied or unusual requests.

Sources

SourceLink
github.comgithub.com/agentscope-ai/agentscope/
github.comgithub.com/agentscope-ai/agentscope/commit/6b43b13a46401e86abd3defeed9a8bfe5b37bc6d
github.comgithub.com/agentscope-ai/agentscope/issues/3313
github.comgithub.com/agentscope-ai/agentscope/pull/2629
github.comgithub.com/agentscope-ai/agentscope/releases/tag/v2.0.9
github.comgithub.com/ybyu-ieu/agent-security-advisories/blob/main/agentscope/agentscope-bash-tool-tee-misclassification-arbitrary-file-write.md
vuldb.comvuldb.com/cve/CVE-2026-123023
vuldb.comvuldb.com/submit/1033131
vuldb.comvuldb.com/vuln/431396
vuldb.comvuldb.com/vuln/431396/cti

Frequently asked

questions

Remote exploitation is reported as possible. Practical exposure depends on whether an attacker can reach and invoke the Bash Tool in a particular deployment.

The available description does not specify authentication requirements. Access to the tool interface and the deployment's access controls determine the relevant prerequisites.

Enforce explicit, fail-closed authorization checks for every sensitive action, and run the tool with only the privileges it needs.

Use centralized, testable authorization policies; review all paths to sensitive operations; and restrict tool access and privileges by default.

Remote exploitation is reported as possible. Practical exposure depends on whether an attacker can reach and invoke the Bash Tool in a particular deployment.

The available description does not specify authentication requirements. Access to the tool interface and the deployment's access controls determine the relevant prerequisites.

Enforce explicit, fail-closed authorization checks for every sensitive action, and run the tool with only the privileges it needs.

Use centralized, testable authorization policies; review all paths to sensitive operations; and restrict tool access and privileges by default.

Find out what's already
exploitable in your codebase.