diff --git a/crates/fractal-kernel/src/config.rs b/crates/fractal-kernel/src/config.rs
index 4ad04741175..7f878a9de72 100644
--- a/crates/fractal-kernel/src/config.rs
+++ b/crates/fractal-kernel/src/config.rs
@@ -469,7 +469,8 @@ impl LimitsConfig {
 #[derive(Clone, Debug, Default, Deserialize, Eq, PartialEq, Serialize)]
 #[serde(default, deny_unknown_fields)]
 pub struct ProgramsConfig {
-    /// `audit` lets a write outside a run's authority proceed and records it; `enforce` refuses it.
+    /// `enforce` (the default) refuses a write outside a run's authority; `audit` lets it proceed
+    /// and records it.
     pub authority: RunAuthorityMode,
 }
 
@@ -480,8 +481,8 @@ impl ProgramsConfig {
 #[derive(Clone, Copy, Debug, Default, Deserialize, Eq, PartialEq, Serialize)]
 #[serde(rename_all = "kebab-case")]
 pub enum RunAuthorityMode {
-    #[default]
     Audit,
+    #[default]
     Enforce,
 }
 
@@ -1414,6 +1415,16 @@ fn machine_host_name() -> String {
 
 #[cfg(test)]
 mod tests {
+    #[test]
+    fn run_authority_is_enforced_unless_the_config_asks_for_audit() {
+        use super::RunAuthorityMode;
+        let absent: super::Config = toml::from_str("").unwrap();
+        assert_eq!(absent.programs.authority, RunAuthorityMode::Enforce);
+        let audit: super::Config = toml::from_str("[programs]\nauthority = \"audit\"\n").unwrap();
+        assert_eq!(audit.programs.authority, RunAuthorityMode::Audit);
+        assert!(toml::from_str::<super::Config>("[programs]\nauthority = \"off\"\n").is_err());
+    }
+
     #[test]
     fn usage_writer_configuration_is_explicit_and_agent_only() {
         let absent: super::Config = toml::from_str("").unwrap();
diff --git a/crates/fractal-store/src/store/run_authority.rs b/crates/fractal-store/src/store/run_authority.rs
index d5bef074d4d..810a175fe4b 100644
--- a/crates/fractal-store/src/store/run_authority.rs
+++ b/crates/fractal-store/src/store/run_authority.rs
@@ -7,8 +7,8 @@
 //! person, agent or daemon starter holds everything in free mode except what the rules restrict,
 //! so each write is also decided by the rules for that starter.
 //!
-//! A write outside the authority is audited (`rule.audited` on `rule/run-authority`) or refused
-//! (`run-authority-denied`), as the daemon's `[programs] authority` says.
+//! A write outside the authority is refused (`run-authority-denied`), or audited (`rule.audited`
+//! on `rule/run-authority`) when the daemon's `[programs] authority` says `audit`.
 use super::*;
 use crate::config::RunAuthorityMode;
 use crate::model::{AuthorityGrant, AuthorityVerb};
@@ -20,7 +20,8 @@ pub const AUDIT_RULE: &str = "run-authority";
 /// Run starters nest at most this deep; a deeper chain grants nothing.
 const MAX_STARTERS: usize = 32;
 
-static ENFORCE: AtomicBool = AtomicBool::new(false);
+/// Enforced until the daemon's config says otherwise, as `[programs] authority` defaults.
+static ENFORCE: AtomicBool = AtomicBool::new(true);
 
 /// Set what a write outside a run's authority does, from the daemon's `[programs] authority`.
 pub fn set_run_authority_mode(mode: RunAuthorityMode) {
diff --git a/crates/fractal-store/src/store/run_authority_tests.rs b/crates/fractal-store/src/store/run_authority_tests.rs
index fa616485878..409907e47f5 100644
--- a/crates/fractal-store/src/store/run_authority_tests.rs
+++ b/crates/fractal-store/src/store/run_authority_tests.rs
@@ -214,7 +214,8 @@ fn a_run_writes_only_within_its_authority_intersected_with_its_starters() {
 #[test]
 fn writers_other_than_runs_are_untouched() {
     let _mode = hold_mode();
-    set_run_authority_mode(RunAuthorityMode::Enforce);
+    // Enforced by default, before any config sets the mode.
+    assert_eq!(run_authority_mode(), RunAuthorityMode::Enforce);
     let store = missions();
     write(&store, "person/operator", "custom/other/person").unwrap();
     store.append_claim(&ClaimInput {
@@ -226,5 +227,5 @@ fn writers_other_than_runs_are_untouched() {
         expected_subject: None,
         idempotency_key: None,
     }).unwrap();
-    set_run_authority_mode(RunAuthorityMode::Audit);
+    assert_eq!(run_authority_mode(), RunAuthorityMode::Enforce);
 }
diff --git a/docs/site/reference/configuration.md b/docs/site/reference/configuration.md
index ba3b7641dae..9246d98a000 100644
--- a/docs/site/reference/configuration.md
+++ b/docs/site/reference/configuration.md
@@ -233,11 +233,12 @@ daemon's `[programs]` table says what a write outside it does:
 
 ```toml
 [programs]
-authority = "audit"   # the default: the write proceeds and `rule.audited` records it on `rule/run-authority`
-# authority = "enforce"   # the write is refused with `run-authority-denied`
+authority = "enforce"   # the default: the write is refused with `run-authority-denied`
+# authority = "audit"    # the write proceeds and `rule.audited` records it on `rule/run-authority`
 ```
 
-Read the audit log with `fractal rules audit run-authority` before you enforce. The setting is read
+Audit lets you discover the grants a new program needs: read them with
+`fractal rules audit run-authority`, declare them, then return to enforce. The setting is read
 when the daemon starts.
 
 ## Definitions do not start work
diff --git a/docs/st3/mission-graph-runtime.md b/docs/st3/mission-graph-runtime.md
index defbd973036..f6eb69699fa 100644
--- a/docs/st3/mission-graph-runtime.md
+++ b/docs/st3/mission-graph-runtime.md
@@ -258,9 +258,9 @@ generation with a predecessor needs `revise ID`, as does everything a revision w
 it changes, each on that run's mission; any other claim needs `write SUBJECT`. A run always
 journals `program.step` on its own step runs. `cancel` is reserved and grants nothing yet: setting
 another run's outcome, which is how a run is cancelled, takes a person or agent actor. A write
-outside the effective authority is recorded as `rule.audited` on `rule/run-authority` and
-proceeds, or is refused with `run-authority-denied`, as the daemon's `[programs] authority` says
-(`audit` by default).
+outside the effective authority is refused with `run-authority-denied`, or, when the daemon's
+`[programs] authority` is `audit`, recorded as `rule.audited` on `rule/run-authority` and let
+through (`enforce` by default).
 
 A request whose caller's ancestry carries a live run credential
 (`ST_RUN_CREDENTIAL`, one per program attempt) is bound to `mission-run/RUN`: like a bound agent, it
diff --git a/release-notes/mission-run-authority-enforced.json b/release-notes/mission-run-authority-enforced.json
new file mode 100644
index 00000000000..981f1409ebf
--- /dev/null
+++ b/release-notes/mission-run-authority-enforced.json
@@ -0,0 +1,11 @@
+{
+  "version": 1,
+  "summary": "Mission run authority is enforced by default: a claim a program writes as `mission-run/RUN` outside its declared authority, intersected with its starter's, is refused with `run-authority-denied`; `[programs] authority = \"audit\"` lets it through and records it instead",
+  "replay": {"status": "none", "detail": "No claim, projection or replay change."},
+  "schema": {"status": "none", "detail": "No table, index or migration change."},
+  "rules": {"status": "changed", "detail": "The default of `[programs] authority` changes from `audit` to `enforce`, both in the config and in the store before the daemon reads its config. Writers that are not mission runs are untouched."},
+  "compatibility": {"status": "changed", "detail": "A program whose mission does not declare the authority its effects use now has those writes refused. Programs land in production only from the next milestone; a daemon that needs to discover grants first sets `[programs] authority = \"audit\"`."},
+  "downtime": {"status": "none", "detail": "The mode is read when the daemon starts; changing it needs a restart."},
+  "manual": {"status": "none", "detail": "No operator action is required for existing work, since nothing wrote as a run before program dispatch carried run credentials."},
+  "recovery": {"status": "none", "detail": "Set `[programs] authority = \"audit\"` and restart to let refused writes proceed while their grants are declared."}
+}