# CVE-2026-96620: Claude Agent SDK argument injection via resume option

Source: https://junglewise.ai/threats/cve-2026-96620-claude-agent-sdk-argument-injection-via-resume-option-allows  
Publisher: Junglewise Threat Intelligence, https://junglewise.ai/threats  
Published: 2026-10-16. Updated: 2026-10-16  
License: CC BY 4.0 with a link back

CVE-2026-96620 · Severity: critical · CVSS 4 · Published 2026-10-16

The Claude Agent SDK for Python, a library used to integrate Claude AI capabilities into applications, passes user-supplied session IDs to the Claude CLI without proper validation. An attacker who can control a session ID can inject malicious command-line flags to execute arbitrary commands on the system, bypassing the SDK's security controls. This could allow an attacker to read sensitive data, modify system files, or gain control of the affected system.

Technologies: [Anthropic Claude Agent SDK](https://junglewise.ai/threats/technologies/claude-agent-sdk). Vendors: [Anthropic](https://junglewise.ai/threats/vendors/anthropic), [PyPI](https://junglewise.ai/threats/vendors/pypi).

## Quick answers

**Is CVE-2026-96620 exploited in the wild?** No. Junglewise Threat Intelligence has no report of CVE-2026-96620 being exploited in the wild.

**How severe is CVE-2026-96620?** Critical severity, with a CVSS score of 4, published 16 October 2026.

**Which products and versions are affected?** Anthropic Claude Agent SDK < 0.2.121; claude-agent-sdk (PyPI) < 0.2.121, ECOSYSTEM:introduced=0,fixed=0.2.121.

**Is there a fix for CVE-2026-96620?** Yes. claude-agent-sdk (PyPI): upgrade to 0.2.121 or later. Junglewise finds every place you run it, drafts the upgrade and applies it once you approve.

**How is CVE-2026-96620 attacked?** It is attacked over the network, in a software library.

**Am I affected by CVE-2026-96620?** Check whether you run Anthropic Claude Agent SDK or claude-agent-sdk (PyPI) at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

## How to fix

- claude-agent-sdk (PyPI): upgrade to 0.2.121 or later

  ```sh
  pip install "claude-agent-sdk>=0.2.121"
  ```


## Technical details

The vulnerability is an argument injection flaw (CWE-88) combined with improper input validation (CWE-20) in the resume parameter passed to the Claude CLI. When a resume value begins with a dash, it is interpreted as a separate CLI flag rather than a session ID. An attacker can inject an --mcp-config flag to define a stdio Model Context Protocol server, achieving arbitrary command execution at CLI startup. Exploitation requires an application to pass untrusted input as the session ID, and the fix is available in version 0.2.121 and later.

## Affected products

- Anthropic claude-agent-sdk < 0.2.121

## Identifiers

- CVE-2026-96620
- GHSA-h4mw-j7qp-8mwm

## Timeline

- 2026-09-12: disclosed
- 2026-09-12: patched: version 0.2.121
- 2026-10-16: advisory

## References

- [https://github.com/anthropics/claude-agent-sdk-python/security/advisories/GHSA-h4mw-j7qp-8mwm](https://github.com/anthropics/claude-agent-sdk-python/security/advisories/GHSA-h4mw-j7qp-8mwm)

## Related threats

- [Anthropic Claude Code OS command injection in authentication helpers](https://junglewise.ai/threats/cve-2026-35022-anthropic-claude-code-os-command-injection-in-authentication)
- [Anthropic Claude Code OS command injection in prompt editor](https://junglewise.ai/threats/cve-2026-35021-anthropic-claude-code-os-command-injection-in-prompt-editor)
- [Anthropic Claude Code CLI OS command injection in command lookup helper](https://junglewise.ai/threats/cve-2026-35020-anthropic-claude-code-cli-os-command-injection-in-command-lookup)