OWASP Agentic
ASI05 · Unexpected Code Execution
Agents generate or run code that escapes the intended sandbox, leading to remote code execution.
OWASP Top 10 for Agentic Applications (2026)
Mitigations
- Sandboxed execution with no ambient credentials
- Egress restrictions
- Review generated code before execution
Detections 3
Related techniques
- LLM05 Improper Output Handling OWASP LLM
In the news
-
MCP server Flyto Core: Flyto2 Core: SSRF guard bypass via DNS rebinding (validate_url_ssrf resolves-then-connects without pinning)
GHSA-6pm8-6f34-9v3g: Flyto Core MCP server's SSRF guard can be bypassed via DNS rebinding because it resolves then connects without pinning.
PatchUpgrade flyto-core to 2.26.8 or later
-
Inside the Codex sandbox
OpenAI describes how the Codex sandbox enforces user-controlled filesystem and network limits when running commands on macOS, Linux, and Windows.
-
External sandboxes for coding agents
Simon Willison and Prime Radiant released three external sandbox server implementations so agent-executed code runs outside the agent, limiting damage from prompt injection.
Tools
Mappings are AI-assisted and checked against the framework definitions; corrections welcome.