diff --git a/src/arbiter/__main__.py b/src/arbiter/__main__.py
index 43169b4..40f053e 100644
--- a/src/arbiter/__main__.py
+++ b/src/arbiter/__main__.py
@@ -25,6 +25,7 @@
 from arbiter.analyzers.duplication_analyzer import DuplicationAnalyzer
 from arbiter.analyzers.ruff_analyzer import RuffAnalyzer
 from arbiter.analyzers.security_analyzer import SecurityAnalyzer
+from arbiter.analyzers.semgrep_analyzer import SemgrepAnalyzer
 from arbiter.diff_analyzer import score_commit, score_diff
 from arbiter.git_historian import count_loc, walk_commits
 from arbiter.scoring import RepoScore, score_findings
@@ -47,6 +48,7 @@ def _get_analyzers() -> list[Analyzer]:
         RuffAnalyzer(),
         ComplexityAnalyzer(),
         SecurityAnalyzer(),
+        SemgrepAnalyzer(),  # opt-in: requires ARBITER_ENABLE_SEMGREP=1
         DeadCodeAnalyzer(),
         DuplicationAnalyzer(),
     ]
diff --git a/src/arbiter/analyzers/semgrep_analyzer.py b/src/arbiter/analyzers/semgrep_analyzer.py
new file mode 100644
index 0000000..a199204
--- /dev/null
+++ b/src/arbiter/analyzers/semgrep_analyzer.py
@@ -0,0 +1,116 @@
+"""Semgrep Analyzer — SAST analysis via semgrep CLI (opt-in).
+
+Gated by the ``ARBITER_ENABLE_SEMGREP`` environment variable. Semgrep is
+disabled by default because rule loading + scanning adds 45-60 seconds to
+``arbiter score`` on typical Python repos, which pushes CI over the 120s
+subprocess timeout used by downstream consumers (e.g. founder-mode's
+``arbiter_adapter.get_quality_digest()``).
+
+Opt in by setting ``ARBITER_ENABLE_SEMGREP=1`` in the environment. When
+enabled, the analyzer uses the curated ``p/security`` ruleset (not
+``--config auto``) to keep runtime predictable.
+
+Reference: regression investigation 2026-04-05 identified an uncommitted
+SemgrepAnalyzer with a 600s timeout and ``--config auto`` as the root
+cause of a 5s → 45-60s ``arbiter score`` slowdown.
+"""
+
+from __future__ import annotations
+
+import json
+import os
+import subprocess
+from pathlib import Path
+
+from arbiter.analyzers.base import Analyzer, Finding
+
+# Env var that enables the analyzer. Disabled by default.
+_ENABLE_ENV = "ARBITER_ENABLE_SEMGREP"
+
+# Curated ruleset — faster than --config auto and predictable scope.
+# Override with ARBITER_SEMGREP_CONFIG=<config> for experimentation.
+_DEFAULT_CONFIG = "p/security"
+
+# Timeout aligned with other analyzers (120s) rather than 600s.
+_TIMEOUT_SECONDS = 120
+
+# Map semgrep severity strings to arbiter's levels.
+_SEVERITY_MAP = {
+    "ERROR": "HIGH",
+    "WARNING": "MEDIUM",
+    "INFO": "LOW",
+    # semgrep also emits uppercase CRITICAL/HIGH/MEDIUM/LOW via some rulesets
+    "CRITICAL": "CRITICAL",
+    "HIGH": "HIGH",
+    "MEDIUM": "MEDIUM",
+    "LOW": "LOW",
+}
+
+
+def _map_severity(raw: str) -> str:
+    return _SEVERITY_MAP.get((raw or "").upper(), "MEDIUM")
+
+
+class SemgrepAnalyzer(Analyzer):
+    """Runs semgrep scan (opt-in) and parses JSON output."""
+
+    @property
+    def name(self) -> str:
+        return "semgrep"
+
+    def is_available(self) -> bool:
+        # Opt-in gate: require explicit env var.
+        if os.environ.get(_ENABLE_ENV, "").strip() not in ("1", "true", "TRUE", "yes"):
+            return False
+        # Tool-present check.
+        try:
+            subprocess.run(
+                ["semgrep", "--version"], capture_output=True, timeout=5,
+            )
+            return True
+        except (FileNotFoundError, subprocess.TimeoutExpired):
+            return False
+
+    def analyze_repo(
+        self, repo_path: Path, exclude_paths: list[str] | None = None,
+    ) -> list[Finding]:
+        config = os.environ.get("ARBITER_SEMGREP_CONFIG", _DEFAULT_CONFIG)
+        cmd = [
+            "semgrep", "scan",
+            "--config", config,
+            "--json",
+            "--quiet",
+            "--metrics", "off",
+        ]
+        for ep in exclude_paths or []:
+            cmd.extend(["--exclude", ep])
+        cmd.append(str(repo_path))
+
+        try:
+            result = subprocess.run(
+                cmd, capture_output=True, text=True, timeout=_TIMEOUT_SECONDS,
+            )
+        except subprocess.TimeoutExpired:
+            return []
+
+        if not result.stdout.strip():
+            return []
+
+        try:
+            data = json.loads(result.stdout)
+        except json.JSONDecodeError:
+            return []
+
+        findings: list[Finding] = []
+        for item in data.get("results", []):
+            extra = item.get("extra", {}) or {}
+            start = item.get("start", {}) or {}
+            findings.append(Finding(
+                file_path=item.get("path", ""),
+                line=start.get("line", 0),
+                severity=_map_severity(extra.get("severity", "")),
+                rule_id=item.get("check_id", ""),
+                message=extra.get("message", ""),
+                tool="semgrep",
+            ))
+        return findings
diff --git a/tests/test_semgrep_analyzer.py b/tests/test_semgrep_analyzer.py
new file mode 100644
index 0000000..7e4504a
--- /dev/null
+++ b/tests/test_semgrep_analyzer.py
@@ -0,0 +1,157 @@
+"""Tests for SemgrepAnalyzer opt-in behavior and JSON parsing."""
+
+import json
+import os
+import subprocess
+from pathlib import Path
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from arbiter.analyzers.semgrep_analyzer import SemgrepAnalyzer
+
+
+@pytest.fixture(autouse=True)
+def _clean_env(monkeypatch):
+    """Ensure each test starts with a known env state."""
+    monkeypatch.delenv("ARBITER_ENABLE_SEMGREP", raising=False)
+    monkeypatch.delenv("ARBITER_SEMGREP_CONFIG", raising=False)
+
+
+class TestSemgrepOptIn:
+    def test_disabled_by_default(self):
+        """Without ARBITER_ENABLE_SEMGREP, is_available() must return False."""
+        analyzer = SemgrepAnalyzer()
+        assert analyzer.is_available() is False
+
+    def test_disabled_when_env_is_false(self, monkeypatch):
+        """ARBITER_ENABLE_SEMGREP=0/false/empty all disable."""
+        for val in ("", "0", "false", "no", "off"):
+            monkeypatch.setenv("ARBITER_ENABLE_SEMGREP", val)
+            assert SemgrepAnalyzer().is_available() is False
+
+    @patch("arbiter.analyzers.semgrep_analyzer.subprocess.run")
+    def test_enabled_when_env_set_and_binary_present(self, mock_run, monkeypatch):
+        """ARBITER_ENABLE_SEMGREP=1 + semgrep binary present → available."""
+        monkeypatch.setenv("ARBITER_ENABLE_SEMGREP", "1")
+        mock_run.return_value = MagicMock(returncode=0)
+        assert SemgrepAnalyzer().is_available() is True
+
+    @patch("arbiter.analyzers.semgrep_analyzer.subprocess.run")
+    def test_disabled_when_env_set_but_binary_missing(self, mock_run, monkeypatch):
+        """Env opt-in alone is not enough — binary must also exist."""
+        monkeypatch.setenv("ARBITER_ENABLE_SEMGREP", "1")
+        mock_run.side_effect = FileNotFoundError()
+        assert SemgrepAnalyzer().is_available() is False
+
+    def test_accepted_truthy_values(self, monkeypatch):
+        """Accept 1, true, TRUE, yes — reject everything else."""
+        with patch(
+            "arbiter.analyzers.semgrep_analyzer.subprocess.run",
+            return_value=MagicMock(returncode=0),
+        ):
+            for val in ("1", "true", "TRUE", "yes"):
+                monkeypatch.setenv("ARBITER_ENABLE_SEMGREP", val)
+                assert SemgrepAnalyzer().is_available() is True, f"val={val!r} should enable"
+
+
+class TestSemgrepParsing:
+    def _sample_semgrep_output(self) -> str:
+        return json.dumps({
+            "results": [
+                {
+                    "check_id": "python.lang.security.audit.eval-detected",
+                    "path": "src/eval_bad.py",
+                    "start": {"line": 10, "col": 1},
+                    "extra": {
+                        "severity": "ERROR",
+                        "message": "eval() use is dangerous",
+                    },
+                },
+                {
+                    "check_id": "python.lang.correctness.unused-import",
+                    "path": "src/utils.py",
+                    "start": {"line": 2, "col": 1},
+                    "extra": {
+                        "severity": "WARNING",
+                        "message": "unused import",
+                    },
+                },
+            ],
+        })
+
+    @patch("arbiter.analyzers.semgrep_analyzer.subprocess.run")
+    def test_parses_json_output(self, mock_run):
+        mock_run.return_value = MagicMock(
+            stdout=self._sample_semgrep_output(), returncode=0,
+        )
+        findings = SemgrepAnalyzer().analyze_repo(Path("/fake/repo"))
+        assert len(findings) == 2
+        assert all(f.tool == "semgrep" for f in findings)
+
+    @patch("arbiter.analyzers.semgrep_analyzer.subprocess.run")
+    def test_severity_mapping(self, mock_run):
+        mock_run.return_value = MagicMock(
+            stdout=self._sample_semgrep_output(), returncode=0,
+        )
+        findings = SemgrepAnalyzer().analyze_repo(Path("/fake"))
+        by_rule = {f.rule_id: f for f in findings}
+        assert by_rule[
+            "python.lang.security.audit.eval-detected"
+        ].severity == "HIGH"
+        assert by_rule[
+            "python.lang.correctness.unused-import"
+        ].severity == "MEDIUM"
+
+    @patch("arbiter.analyzers.semgrep_analyzer.subprocess.run")
+    def test_handles_empty_output(self, mock_run):
+        mock_run.return_value = MagicMock(stdout="", returncode=0)
+        assert SemgrepAnalyzer().analyze_repo(Path("/fake")) == []
+
+    @patch("arbiter.analyzers.semgrep_analyzer.subprocess.run")
+    def test_handles_invalid_json(self, mock_run):
+        mock_run.return_value = MagicMock(stdout="not json {{{", returncode=0)
+        assert SemgrepAnalyzer().analyze_repo(Path("/fake")) == []
+
+    @patch("arbiter.analyzers.semgrep_analyzer.subprocess.run")
+    def test_handles_timeout(self, mock_run):
+        mock_run.side_effect = subprocess.TimeoutExpired(cmd="semgrep", timeout=120)
+        assert SemgrepAnalyzer().analyze_repo(Path("/fake")) == []
+
+    @patch("arbiter.analyzers.semgrep_analyzer.subprocess.run")
+    def test_timeout_is_120s_not_600s(self, mock_run):
+        """Regression guard: timeout must stay at 120s to fit CI contract."""
+        mock_run.return_value = MagicMock(stdout="", returncode=0)
+        SemgrepAnalyzer().analyze_repo(Path("/fake"))
+        _, kwargs = mock_run.call_args
+        assert kwargs.get("timeout") == 120
+
+    @patch("arbiter.analyzers.semgrep_analyzer.subprocess.run")
+    def test_uses_curated_ruleset_by_default(self, mock_run):
+        """Regression guard: don't use --config auto, it's too slow."""
+        mock_run.return_value = MagicMock(stdout="", returncode=0)
+        SemgrepAnalyzer().analyze_repo(Path("/fake"))
+        cmd = mock_run.call_args[0][0]
+        assert "--config" in cmd
+        config_idx = cmd.index("--config")
+        assert cmd[config_idx + 1] == "p/security"
+        assert "auto" not in cmd
+
+    @patch("arbiter.analyzers.semgrep_analyzer.subprocess.run")
+    def test_config_override_via_env(self, mock_run, monkeypatch):
+        monkeypatch.setenv("ARBITER_SEMGREP_CONFIG", "p/python")
+        mock_run.return_value = MagicMock(stdout="", returncode=0)
+        SemgrepAnalyzer().analyze_repo(Path("/fake"))
+        cmd = mock_run.call_args[0][0]
+        config_idx = cmd.index("--config")
+        assert cmd[config_idx + 1] == "p/python"
+
+    @patch("arbiter.analyzers.semgrep_analyzer.subprocess.run")
+    def test_passes_exclude_paths(self, mock_run):
+        mock_run.return_value = MagicMock(stdout="", returncode=0)
+        SemgrepAnalyzer().analyze_repo(
+            Path("/fake"), exclude_paths=["tests/*", "build/*"],
+        )
+        cmd = mock_run.call_args[0][0]
+        assert "tests/*" in cmd
+        assert "build/*" in cmd