fix: redact code evidence in scanner findings (#1495)

* fix: redact code evidence in scanner findings

* fix: harden serialized code evidence redaction

* fix: harden scanner evidence redaction

* fix: fail closed on framed evidence expressions

* fix: handle nul-framed evidence on python 3.10

* fix: preserve file URL evidence paths

* fix: close evidence assignment redaction gaps

* fix: close evidence redaction gaps

* fix: harden keyed evidence redaction

* fix: cover remaining evidence credential channels

* fix: redact literal credential pairs

* fix: close remaining code evidence leaks

* fix: close final evidence redaction gaps

* fix: harden bounded evidence redaction

Preserve safe context across preview boundaries, close indented and structured credential leaks, and keep adversarial redaction scans bounded without hiding benign code.

* fix: harden code evidence redaction

* fix: close code evidence redaction gaps

* fix: preserve redacted executable evidence

* fix: close evidence redaction follow-up gaps

* fix: redact percent-formatted evidence

Author: mldangelo-oai

CHANGELOG.md

@@ -69,6 +69,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - redact secret previews and URL path credentials from metadata scanner findings
 - redact secret-shaped dictionary keys from embedded-secret detector finding contexts
 - redact compound credential names and malformed userinfo URLs in scanner evidence
+- redact secret-bearing JAX/Orbax, JIT, PyTorch ZIP, and explicit model-network evidence before serializing findings
 - restrict JFrog credential forwarding to explicitly trusted HTTPS hosts
 - include content-routed renamed JFrog folder artifacts in selective downloads with fail-closed bounded probes
 - strip JFrog credentials from untrusted redirect hops during artifact and Storage API requests

modelaudit/detectors/jit_script.py

@@ -51,6 +51,13 @@ def create_jit_finding(**kwargs: Any) -> "JITScriptFinding":
     return JITScriptFinding(**kwargs)
 
 
+def _redact_code_evidence_snippet(code: str, max_chars: int = 200) -> str:
+    """Redact credentials from detector code evidence before serializing it."""
+    from modelaudit.scanners._evidence_redaction import redact_evidence_string
+
+    return redact_evidence_string(code, max_chars=max_chars)
+
+
 # Dangerous TorchScript operations that can execute arbitrary code
 DANGEROUS_TORCH_OPS = [
     # System operations
@@ -15248,7 +15255,7 @@ def _extract_and_check_python_code(
                                 recommendation=f"Remove {dangerous_import} import - it can be used maliciously",
                                 confidence=0.9,
                                 framework=framework,
-                                code_snippet=code_str[:200],
+                                code_snippet=_redact_code_evidence_snippet(code_str),
                                 type="dangerous_import",
                                 operation=None,
                                 builtin=None,
@@ -15273,7 +15280,7 @@ def _extract_and_check_python_code(
                                 recommendation=f"Remove {builtin} usage - it can execute arbitrary code",
                                 confidence=0.9,
                                 framework=framework,
-                                code_snippet=code_str[:200],
+                                code_snippet=_redact_code_evidence_snippet(code_str),
                                 type="dangerous_builtin",
                                 operation=None,
                                 builtin=builtin,
@@ -15318,7 +15325,7 @@ def _extract_and_check_python_code(
                     recommendation=f"Remove {builtin} usage - it can execute arbitrary code",
                     confidence=0.9,
                     framework=framework,
-                    code_snippet=code_str[:200],
+                    code_snippet=_redact_code_evidence_snippet(code_str),
                     type="dangerous_builtin",
                     operation=None,
                     builtin=builtin,