Tools AI Risk Radar ai-incident-0232
Incident record
Claude Agent SDK resume input could execute commands
- Severity
- Critical
- Status
- Patched
- Type
- Infra Vuln
- Target
- Claude Agent SDK for Python
- Actor
- researcher
- CVE
- CVE-2026-96620
What happened
An application that passed an attacker-controlled resume value to the SDK could start an attacker-defined MCP command through injected command-line options.
Anthropic’s advisory says the resume value could be interpreted as a CLI flag and bypass the SDK’s permission controls at startup. The affected package is claude-agent-sdk before 0.2.121. This entry records the public advisory date and released fix.
Sources
- Maintainer security advisorygithub.com/anthropics/claude-agent-sdk-python/security/advis…
- GitHub reviewed advisorygithub.com/advisories/GHSA-h4mw-j7qp-8mwm
One record from the AI Risk Radar, maintained by the Security Desk.Data: CSV · JSON ·RSS · CC BY 4.0 with attribution to YFarmX.Tracker updated · 28 October 2026