Read the docs
TI-2026-75555 Aliases CVE-2026-96620GHSA-h4mw-j7qp-8mwm

Argument injection in Claude Agent SDK lets attackers run commands

Priority MEDIUM 49CWE-20Published · Updated

Key facts

Attack vector
Network
Privileges required
None
User interaction
None
Attack complexity
Low
Attack requirements
Present
Impact type
Code execution
Confidentiality · integrity · availability
High · High · High
Fix status
Fixed

Attack profile from CVSS 4.0 · CNA. Fix status from CNA and GitHub. Impact type from CWE-78.

Summary

The Claude Agent SDK for Python before 0.2.121 passes the resume option to the Claude CLI without validation, so a value starting with a dash is parsed as a CLI flag (CWE-88). When an application passes untrusted input as the session ID, an unauthenticated remote attacker can inject the --mcp-config flag to define a stdio MCP server that runs arbitrary commands at CLI startup, outside the SDK permission controls.

Written with Claude Haiku 5.5 from the sources below · 2026-10-28

Description from the CNA

Claude Agent SDK is a Python SDK for Claude Code. Prior to 0.2.121, the resume option is passed to the Claude CLI as a separate token without validation, so a dash-leading value is parsed as an independent CLI flag. An attacker can inject the --mcp-config flag to define a stdio MCP server, causing arbitrary command execution at CLI startup outside the SDK permission controls. The issue is exploitable when an application passes untrusted input as the session ID to resume. This issue is fixed in version 0.2.121.

Why it matters now

Exploitation None reported CISA SSVC · not in CISA KEV
EPSS 0.00705 53rd percentileScored 2026-11-07
Fix Fixed claude-agent-sdk-python (anthropics): 0.2.121claude-agent-sdk (PyPI): 0.2.121

SSVC decision points

CISA ADP · SSVC 2.0.3

Exploitation
None
Automatable
Yes
Technical impact
Total

The decision (Track, Attend, Act) also depends on your mission and well-being impact; see CISA's SSVC guide.

Affected products

VendorProductAffected versionsFixed inStatusSource
anthropicsclaude-agent-sdk-pythonbefore 0.2.1210.2.121Patch availableCNA
—claude-agent-sdk (PyPI)before 0.2.1210.2.121Patch availableGHSA-h4mw-j7qp-8mwm
  • claude-agent-sdk-python Vendoranthropics Affected before 0.2.121 Fixed in 0.2.121 Patch available CNA
  • claude-agent-sdk (PyPI) Vendor— Affected before 0.2.121 Fixed in 0.2.121 Patch available GHSA-h4mw-j7qp-8mwm

Remediation

Upgrade the claude-agent-sdk package on PyPI to version 0.2.121, which fixes the issue. The sources list no fixed version for claude-agent-sdk-python.

Written with Claude Haiku 5.5 from the sources below · 2026-10-28

  • Fix · CNA claude-agent-sdk-python (anthropics) Fixed in 0.2.121
  • Fix · GHSA-h4mw-j7qp-8mwm claude-agent-sdk (PyPI) Fixed in 0.2.121 Advisory →

Timeline

  1. 2026-10-18 Threatist record created from CVE-2026-96620
  2. 2026-10-18 Alias added: GHSA-h4mw-j7qp-8mwm
  3. 2026-10-18 Fix available: PyPI/[email protected] · GHSA-h4mw-j7qp-8mwm
  4. 2026-10-18 CVSS score added · GHSA-h4mw-j7qp-8mwm
  5. 2026-10-18 Weakness changed · GHSA-h4mw-j7qp-8mwm
  6. 2026-10-18 5 new references · GHSA-h4mw-j7qp-8mwm
  7. 2026-10-21 CISA SSVC decision points changed
  8. 2026-10-25 EPSS moved to the 50th–90th percentile

References

Other 5

  • TI-2026-79288 Out-of-bounds write in Google Brotli decoder KEVCVSS 4.0CRITICAL 9.2
  • TI-2026-00421 MCP TypeScript SDK UriTemplate Exploded Array Pattern ReDoS CVSS 4.0HIGH 8.7
  • TI-2025-40558 Claude Code Command Validation Bypass Allows Arbitrary Code Execution CVSS 4.0HIGH 8.7
  • TI-2025-36683 Claude Code vulnerable to arbitrary code execution caused by maliciously configured git email CVSS 4.0HIGH 8.7
  • TI-2025-24240 Anthropic Slack MCP Server Data Exfiltration via Link Unfurling CVSS 4.0CRITICAL 9.3
  • TI-2025-32918 Claude Code IDE extensions allow websocket connections from arbitrary origins CVSS 4.0HIGH 8.8
See all 46 →

Sources

This product uses the NVD API but is not endorsed or certified by the NVD. Sources and licences