Affected Systems
Multiple AI development tools: Cursor IDE, OpenAI Codex, Google Gemini CLI, and Antigravity. Vulnerability affects AI agent sandbox implementations where agents write files executed by host tools, allowing escape from restricted environments.
Exploitation Status
Research disclosure. No CVE assigned yet. Exploitation status unclear—likely proof-of-concept demonstrated by researchers. No evidence of active in-the-wild exploitation reported.
Business Impact
AI-assisted development environments may allow malicious or compromised AI agents to execute arbitrary code on developer workstations, bypassing sandbox protections. Risk is elevated in environments where developers use AI coding assistants with access to sensitive codebases, credentials, or internal networks. No CVSS score published. Impact depends on deployment context and trust boundaries between AI agents and host systems.
Urgency
🟡 Within a week
Recommended Actions
- Inventory all AI development tools in use (Cursor, Codex integrations, Gemini CLI, Antigravity) and identify affected versions.
- Monitor vendor security advisories from Cursor, OpenAI, Google, and Antigravity for patches and updated sandbox implementations.
- Review and restrict file execution permissions for directories where AI agents write output files.
- Implement application allowlisting or execution policies to prevent unauthorized code execution from AI agent workspace directories.
- Educate developers on risks of AI-generated code execution and enforce code review before running AI-suggested scripts or commands.