diff --git a/docs/trust-boundary/security.md b/docs/trust-boundary/security.md
index c6fa8982..ef6c7a75 100644
--- a/docs/trust-boundary/security.md
+++ b/docs/trust-boundary/security.md
@@ -45,6 +45,28 @@ matrix **identities** — the public handles (`name`, `user_id` `@user:server`,
 peers can find and address one another on a shared matrix instance. Only the
 public handle crosses that boundary; the token never does.
 
+### The swarm secret store isn't a boundary between hives
+
+A hive reads its agents' credentials out of the swarm's secret store with its
+own certificate. **Every hive's policy grants read on every agent's
+credentials**, not only on the agents it hosts — so a compromised hive can read
+the matrix token of an agent running on a different hive.
+
+That's deliberate and it's the interim state, not the intent. An agent's
+credential path doesn't name the hive hosting it (agents move between hives),
+so a per-hive grant has to be an enumeration the controller re-emits whenever
+the roster changes — and an enumeration that can drift or land out of order
+advertises a boundary it doesn't actually hold. A wide grant that says what it
+is beats a narrow one that only looks narrow.
+
+What still holds: the grant is **read-only** (a hive can't write an agent's
+credential, so it can't hand itself an agent's identity), and it's scoped to
+the agent-credential prefix — nothing else in the store is reachable with it.
+
+**#4137** tracks narrowing it, with the two candidate directions: scope
+the grant per hive (and pay for the re-emission), or give each agent container
+its own store identity so credentials never pass through a hive at all.
+
 ### Threat model: prompt injection → confused deputy
 
 The realistic adversary **never needs to breach the container**. They supply
diff --git a/swarm-controller/src/main.rs b/swarm-controller/src/main.rs
index b7ca8b75..00454f5b 100644
--- a/swarm-controller/src/main.rs
+++ b/swarm-controller/src/main.rs
@@ -47,7 +47,9 @@ mod forge;
 mod issue_report;
 mod matrix_account;
 mod otel_http_client;
+mod read_policy;
 mod status;
+mod store;
 mod vcs_metrics;
 mod wanted;
 mod webhook;
@@ -1620,8 +1622,15 @@ async fn main() -> Result<()> {
     let config_prs = forge_client.clone().map(config_pr::spawn);
     let state_forge = keep_forge_for_state(forge_client, webhook_secret.clone());
 
+    let hives = load_hives();
+    // Before serving, because a hive whose policy does not exist cannot read
+    // anything this daemon writes for it. Same "log and carry on" shape as
+    // every connect above.
+    read_policy::ensure_hive_policies(&hives.iter().map(|h| h.name.clone()).collect::<Vec<_>>())
+        .await;
+
     let state = AppState {
-        hives: Arc::new(load_hives()),
+        hives: Arc::new(hives),
         links: Arc::new(load_links()),
         wanted: wanted_writer(status.as_ref()),
         agent_status: agent_status_reader(status.as_ref()),
diff --git a/swarm-controller/src/matrix_account.rs b/swarm-controller/src/matrix_account.rs
index 31843bda..46f8d38c 100644
--- a/swarm-controller/src/matrix_account.rs
+++ b/swarm-controller/src/matrix_account.rs
@@ -26,20 +26,11 @@ use axum::extract::State;
 use axum::http::StatusCode;
 use serde::{Deserialize, Serialize};
 use swarm_queue_client::{CredentialNotice, credential_subject};
-use swarm_secret_client::{SecretStore, matrix};
+use swarm_secret_client::matrix;
 use utoipa::ToSchema;
 
 use super::{AppState, error_problem, swarm_hive};
 
-/// The cert-auth role this daemon logs into the secret store as.
-///
-/// `nix/host-modules/swarm-bao.nix`'s `controllerPolicyName` creates the role,
-/// names the policy after it, and `nix/module-eval.nix` pins the literal.
-///
-/// ⚠️ Not the certificate's CN. The role *matches on* the CN
-/// (`allowed_common_names`), so the two are deliberately different strings.
-const CERT_ROLE: &str = "swarm-controller";
-
 fn default_mode() -> String {
     "token".to_owned()
 }
@@ -156,7 +147,7 @@ pub async fn put_matrix_account(
     // touched, so a failed login leaves no partial state behind.
     let (token, homeserver, user_id) = resolve_credential(&req).await.map_err(|b| *b)?;
 
-    let store = SecretStore::from_env(CERT_ROLE).await.map_err(|e| {
+    let store = crate::store::connect().await.map_err(|e| {
         tracing::warn!(error = %e, "connecting to the swarm secret store failed");
         error_problem(StatusCode::INTERNAL_SERVER_ERROR, &e.to_string())
     })?;
diff --git a/swarm-controller/src/read_policy.rs b/swarm-controller/src/read_policy.rs
new file mode 100644
index 00000000..f00297e4
--- /dev/null
+++ b/swarm-controller/src/read_policy.rs
@@ -0,0 +1,48 @@
+//! Writing the policy a hive's own certificate logs in with.
+//!
+//! The controller writes a credential; the hive fetches it with its own
+//! certificate. Nothing said which paths that certificate may read, so the
+//! read half of every delivery answers 403.
+//!
+//! A startup pass, not a hook: the document is the same for every hive and
+//! does not depend on which agents exist ([`swarm_secret_client::policy`]
+//! explains why it is that wide), so there is nothing to keep in step with
+//! anything. Per-hive failures are reported and skipped — one unwritable
+//! policy should not take down a daemon that serves everything else, and the
+//! next start retries it.
+
+use anyhow::{Context, Result};
+use swarm_secret_client::{SecretStore, policy};
+
+/// Give every hive in `hives` the read policy its certificate will carry.
+///
+/// Does nothing when this deployment has no store identity, which is the
+/// shape a controller without a secret store runs in.
+pub async fn ensure_hive_policies(hives: &[String]) {
+    if hives.is_empty() {
+        return;
+    }
+    let store = match crate::store::connect().await {
+        Ok(store) => store,
+        Err(e) => {
+            tracing::info!(reason = %e, "no secret store reachable; hive read policies are not managed here");
+            return;
+        }
+    };
+    for hive in hives {
+        if let Err(e) = write_one(&store, hive).await {
+            tracing::warn!(hive, error = %format!("{e:#}"), "writing this hive's read policy failed");
+        }
+    }
+}
+
+/// One hive's policy: its own name, the shared document.
+async fn write_one(store: &SecretStore, hive: &str) -> Result<()> {
+    let name = policy::hive_object_name(hive)?;
+    store
+        .write_policy(&name, &policy::render())
+        .await
+        .with_context(|| format!("writing the read policy {name}"))?;
+    tracing::info!(hive, policy = %name, "hive read policy in place");
+    Ok(())
+}
diff --git a/swarm-controller/src/store.rs b/swarm-controller/src/store.rs
new file mode 100644
index 00000000..d7875876
--- /dev/null
+++ b/swarm-controller/src/store.rs
@@ -0,0 +1,25 @@
+//! The controller's own identity at the swarm's secret store.
+//!
+//! Two paths log in: writing an agent's credential, and writing the read
+//! grant that lets a hive fetch one back. Both present the same certificate
+//! under the same role, so the role is named here rather than at each caller.
+
+use swarm_secret_client::{Error, SecretStore};
+
+/// The cert-auth role the controller logs in under.
+///
+/// `nix/host-modules/swarm-bao.nix`'s `controllerPolicyName` creates the role,
+/// names the policy after it, and `nix/module-eval.nix` pins the literal.
+///
+/// ⚠️ Not the certificate's CN. The role *matches on* the CN
+/// (`allowed_common_names`), so the two are deliberately different strings.
+pub const CERT_ROLE: &str = "swarm-controller";
+
+/// Log in to the store with this deployment's certificate.
+///
+/// # Errors
+/// Whatever [`SecretStore::from_env`] raises — an unset `BAO_*` variable, an
+/// unreadable identity file, or a store that refuses the login.
+pub async fn connect() -> Result<SecretStore, Error> {
+    SecretStore::from_env(CERT_ROLE).await
+}
diff --git a/swarm-queue-client/src/wanted.rs b/swarm-queue-client/src/wanted.rs
index 98140b1e..080032a7 100644
--- a/swarm-queue-client/src/wanted.rs
+++ b/swarm-queue-client/src/wanted.rs
@@ -92,9 +92,12 @@ pub enum AgentState {
     /// Exists on the hive and is not running.
     Offline,
     /// Torn down entirely — the hive runs the destroy template once, then
-    /// this agent drops out of the declared set rather than staying as a
-    /// third steady state. Irreversible: there is no state that brings a
-    /// destroyed agent back short of a fresh deploy.
+    /// treats absence as agreement rather than re-running it. The key
+    /// stays in the declared set with this state forever, on purpose: it's
+    /// what lets a redeclare of an already-destroyed agent be rejected as
+    /// a terminal-state transition instead of silently reviving it.
+    /// Irreversible: there is no state that brings a destroyed agent back
+    /// short of a fresh deploy.
     Destroyed,
 }
 
diff --git a/swarm-secret-client/src/client.rs b/swarm-secret-client/src/client.rs
index 74e61af8..fcd9bda1 100644
--- a/swarm-secret-client/src/client.rs
+++ b/swarm-secret-client/src/client.rs
@@ -151,6 +151,22 @@ impl SecretStore {
         vaultrs::kv2::set(&self.inner, MOUNT, path, value).await?;
         Ok(())
     }
+
+    /// Replace the ACL policy named `name` with `policy`.
+    ///
+    /// A whole-document write, not a merge: the store has no other verb, and
+    /// the caller renders the document from the current agent set anyway, so
+    /// a stanza that is gone from the render is meant to be gone from the
+    /// grant. Render the text with [`crate::policy`] rather than by hand.
+    ///
+    /// # Errors
+    /// [`Error::Vault`] when the token's own policy does not cover
+    /// `sys/policies/acl/<name>` — which is what a controller scoped to the
+    /// `hive-*` namespace gets for any other name.
+    pub async fn write_policy(&self, name: &str, policy: &str) -> Result<(), Error> {
+        vaultrs::sys::policy::set(&self.inner, name, policy).await?;
+        Ok(())
+    }
 }
 
 #[cfg(test)]
diff --git a/swarm-secret-client/src/lib.rs b/swarm-secret-client/src/lib.rs
index a146e0cd..d54d5d68 100644
--- a/swarm-secret-client/src/lib.rs
+++ b/swarm-secret-client/src/lib.rs
@@ -8,6 +8,11 @@
 //! ([`matrix`]). Each of those is a thing the controller and a hive must say
 //! identically, so it is said once here.
 //!
+//! [`policy`] is the same kind of agreement seen from the other side: which of
+//! those paths a hive's own token may read. It belongs here rather than in the
+//! controller because the grant and the path are one statement — spelled
+//! differently they produce a 403 that names neither.
+//!
 //! [`client`] is deliberately ignorant of all of it: it moves whatever type a
 //! caller names, so a second kind of secret is a new module beside [`matrix`]
 //! and not another field on a struct shared with it.
@@ -15,6 +20,7 @@
 pub mod client;
 pub mod matrix;
 pub mod path;
+pub mod policy;
 
 pub use client::SecretStore;
 
diff --git a/swarm-secret-client/src/policy.rs b/swarm-secret-client/src/policy.rs
new file mode 100644
index 00000000..de45771e
--- /dev/null
+++ b/swarm-secret-client/src/policy.rs
@@ -0,0 +1,115 @@
+//! The read agreement: which credentials a hive's own token may fetch.
+//!
+//! The mirror of [`crate::matrix`]. That module says where a credential lives;
+//! this one says who is allowed to read it, and the two have to agree on the
+//! same path or a delivery fails with a 403 that names nothing.
+//!
+//! ⚠️ Every hive gets the same document, and it grants read on **every**
+//! agent's credentials rather than on the ones that hive hosts. That is a
+//! decision, not an oversight: an agent's path does not name its hive, so a
+//! per-hive grant has to be enumerated and re-emitted, and an enumeration that
+//! can silently drift advertises a boundary it does not hold. A wide grant that
+//! says so beats a narrow one that only looks narrow. The narrower shapes, and
+//! what they would cost, are in `docs/trust-boundary/security.md`.
+//!
+//! Rendering stays separate from writing so the text can be asserted with no
+//! store to talk to.
+
+use crate::{
+    Error,
+    path::{AGENT_PREFIX, MOUNT, checked_segment},
+};
+
+/// Namespace for a hive's own policy and cert-auth role.
+///
+/// The controller's own grant is scoped to `hive-*` for both, so this prefix is
+/// the difference between a hive the controller may provision and a policy it
+/// must not be able to rewrite — including its own.
+pub const HIVE_PREFIX: &str = "hive-";
+
+/// The policy and cert-auth role name for `hive`. One name, both objects: the
+/// role attaches the policy by spelling it identically.
+///
+/// # Errors
+/// [`Error::PathSegment`] when `hive` holds anything but `[A-Za-z0-9_-]`.
+pub fn hive_object_name(hive: &str) -> Result<String, Error> {
+    checked_segment("hive", hive)?;
+    Ok(format!("{HIVE_PREFIX}{hive}"))
+}
+
+/// Render the document every hive's policy holds: read on every agent's
+/// credentials.
+///
+/// Takes no arguments because it depends on nothing — same text for every
+/// hive, unchanged by which agents exist. That is what makes it a deploy-time
+/// object rather than derived state with a re-emission to get wrong.
+///
+/// Read-only: the controller mints these and never reads one back.
+#[must_use]
+pub fn render() -> String {
+    format!("path \"{MOUNT}/data/{AGENT_PREFIX}/*\" {{\n  capabilities = [\"read\"]\n}}\n")
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn the_document_grants_read_over_the_whole_agent_prefix() {
+        assert_eq!(
+            render(),
+            "path \"secret/data/swarm/agents/*\" {\n  capabilities = [\"read\"]\n}\n"
+        );
+    }
+
+    #[test]
+    fn the_grant_is_read_only() {
+        // The half of the old policy that survives the widening: a hive reads
+        // credentials, and a hive that could write one could hand itself an
+        // agent's identity.
+        let p = render();
+        assert!(!p.contains("create"));
+        assert!(!p.contains("update"));
+        assert!(!p.contains("delete"));
+        assert!(!p.contains("list"));
+    }
+
+    #[test]
+    fn no_name_reaches_the_document_at_all() {
+        // Why the injection cases that used to live here are gone rather than
+        // relaxed: nothing interpolates into the text any more, so there is no
+        // stanza for a name to close. `hive_object_name` still validates,
+        // because a name does reach the policy's *identifier*.
+        let p = render();
+        assert!(!p.contains("pr1ma"));
+        assert_eq!(p.matches("path \"").count(), 1, "one stanza, no per-agent");
+        assert!(
+            hive_object_name("atlas/*\" { capabilities = [\"root\"] }").is_err(),
+            "the object NAME is still a place a name can do damage"
+        );
+    }
+
+    #[test]
+    fn the_legal_charset_is_actually_reachable() {
+        // The control for the case above: if every name were refused, that
+        // assertion would pass while proving nothing.
+        assert!(hive_object_name("a-b_C9").is_ok());
+    }
+
+    #[test]
+    fn every_hive_gets_a_byte_identical_document() {
+        // The property the deploy-time write depends on: nothing about a hive
+        // or its agents changes the text, so there is nothing to re-emit.
+        assert_eq!(render(), render());
+    }
+
+    #[test]
+    fn the_object_name_sits_inside_the_namespace_the_controller_may_write() {
+        // `hive-` is what the controller's own policy scopes both
+        // `sys/policies/acl/` and `auth/cert/certs/` to, so a name outside it
+        // is one the controller cannot create at all.
+        let n = hive_object_name("pr1ma").expect("legal");
+        assert_eq!(n, "hive-pr1ma");
+        assert!(n.starts_with(HIVE_PREFIX));
+    }
+}