Security

SECURITY.md

Security Model

SIFT Sentinel is not a malware sandbox. It is an evidence-safe orchestration layer for DFIR tooling.

Controls

No generic shell command is exposed through MCP.
Evidence root paths are read-only by enforced policy.
Writes outside the configured output root are denied.
SIFT subprocesses use allowlisted binaries and argv arrays.
Volatility plugins are allowlisted.
Tool output is structured before it reaches the model.
Confirmed findings require evidence references and tool call IDs.

Reporting Issues

Open a GitHub issue with:

Tool name
Case configuration
Expected behavior
Observed behavior
Whether evidence integrity was affected

There aren't any published security advisories