diff --git a/hyperactor/src/config.rs b/hyperactor/src/config.rs
index e6f43b2b6..484c9ea76 100644
--- a/hyperactor/src/config.rs
+++ b/hyperactor/src/config.rs
@@ -40,9 +40,6 @@ declare_attrs! {
 
     /// Maximum buffer size for split port messages
     pub attr SPLIT_MAX_BUFFER_SIZE: usize = 5;
-
-    /// Flag indicating if this is a managed subprocess
-    pub attr IS_MANAGED_SUBPROCESS: bool = false;
 }
 
 /// Load configuration from environment variables
@@ -84,8 +81,6 @@ pub fn from_env() -> Attrs {
         }
     }
 
-    config[IS_MANAGED_SUBPROCESS] = env::var("HYPERACTOR_MANAGED_SUBPROCESS").is_ok();
-
     config
 }
 
@@ -121,9 +116,6 @@ pub fn merge(config: &mut Attrs, other: &Attrs) {
     if other.contains_key(SPLIT_MAX_BUFFER_SIZE) {
         config[SPLIT_MAX_BUFFER_SIZE] = other[SPLIT_MAX_BUFFER_SIZE];
     }
-    if other.contains_key(IS_MANAGED_SUBPROCESS) {
-        config[IS_MANAGED_SUBPROCESS] = other[IS_MANAGED_SUBPROCESS];
-    }
 }
 
 /// Global configuration functions
@@ -294,7 +286,6 @@ mod tests {
         );
         assert_eq!(config[MESSAGE_ACK_EVERY_N_MESSAGES], 1000);
         assert_eq!(config[SPLIT_MAX_BUFFER_SIZE], 5);
-        assert!(!config[IS_MANAGED_SUBPROCESS]);
     }
 
     #[test]
@@ -375,7 +366,6 @@ mod tests {
         );
         assert_eq!(config[MESSAGE_ACK_EVERY_N_MESSAGES], 1000);
         assert_eq!(config[SPLIT_MAX_BUFFER_SIZE], 5);
-        assert!(!config[IS_MANAGED_SUBPROCESS]);
 
         // Verify the keys have defaults
         assert!(CODEC_MAX_FRAME_LENGTH.has_default());
@@ -383,7 +373,6 @@ mod tests {
         assert!(MESSAGE_ACK_TIME_INTERVAL.has_default());
         assert!(MESSAGE_ACK_EVERY_N_MESSAGES.has_default());
         assert!(SPLIT_MAX_BUFFER_SIZE.has_default());
-        assert!(IS_MANAGED_SUBPROCESS.has_default());
 
         // Verify we can get defaults directly from keys
         assert_eq!(
@@ -400,7 +389,6 @@ mod tests {
         );
         assert_eq!(MESSAGE_ACK_EVERY_N_MESSAGES.default(), Some(&1000));
         assert_eq!(SPLIT_MAX_BUFFER_SIZE.default(), Some(&5));
-        assert_eq!(IS_MANAGED_SUBPROCESS.default(), Some(&false));
     }
 
     #[test]
diff --git a/hyperactor/src/init.rs b/hyperactor/src/init.rs
index ef4f96ffe..0ba057f1d 100644
--- a/hyperactor/src/init.rs
+++ b/hyperactor/src/init.rs
@@ -22,9 +22,6 @@ pub(crate) static RUNTIME: LazyLock<tokio::runtime::Runtime> =
 /// Initialize the Hyperactor runtime. Specifically:
 /// - Set up panic handling, so that we get consistent panic stack traces in Actors.
 /// - Initialize logging defaults.
-/// - On Linux, set up signal handlers to ensure that managed child processes are reliably
-///   terminated when their parents die. This is indicated by the environment variable
-///   `HYPERACTOR_MANAGED_SUBPROCESS`.
 pub fn initialize() {
     static INITIALIZED: OnceLock<()> = OnceLock::new();
     INITIALIZED.get_or_init(|| {
@@ -38,15 +35,8 @@ pub fn initialize() {
 #[cfg(target_os = "linux")]
 mod linux {
     use std::backtrace::Backtrace;
-    use std::process;
 
-    use libc::PR_SET_PDEATHSIG;
-    use nix::sys::signal::SIGUSR1;
     use nix::sys::signal::SigHandler;
-    use nix::unistd::getpid;
-    use nix::unistd::getppid;
-    use tokio::signal::unix::SignalKind;
-    use tokio::signal::unix::signal;
 
     pub(crate) fn initialize() {
         // Safety: Because I want to
@@ -68,29 +58,5 @@ mod linux {
             )
             .expect("unable to register signal handler");
         }
-
-        if !crate::config::global::get(crate::config::IS_MANAGED_SUBPROCESS) {
-            return;
-        }
-        super::RUNTIME.spawn(async {
-            match signal(SignalKind::user_defined1()) {
-                Ok(mut sigusr1) => {
-                    // SAFETY: required for signal handling
-                    unsafe {
-                        libc::prctl(PR_SET_PDEATHSIG, SIGUSR1);
-                    }
-                    sigusr1.recv().await;
-                    tracing::error!(
-                        "hyperactor[{}]: parent process {} died; exiting",
-                        getpid(),
-                        getppid()
-                    );
-                    process::exit(1);
-                }
-                Err(err) => {
-                    eprintln!("failed to set up SIGUSR1 signal handler: {:?}", err);
-                }
-            }
-        });
     }
 }
diff --git a/monarch_tensor_worker/src/lib.rs b/monarch_tensor_worker/src/lib.rs
index a4236deee..1e507f770 100644
--- a/monarch_tensor_worker/src/lib.rs
+++ b/monarch_tensor_worker/src/lib.rs
@@ -2381,7 +2381,6 @@ mod tests {
                 .arg(format!("--bootstrap-addr={system_addr}"))
                 .arg(format!("--world-id={world_id}"))
                 .arg(format!("--proc-id={proc_id}"))
-                .env("HYPERACTOR_MANAGED_SUBPROCESS", "1")
                 .stdout(Stdio::piped())
                 .stdin(Stdio::piped())
                 .kill_on_drop(true)
diff --git a/python/monarch/mesh_controller.py b/python/monarch/mesh_controller.py
index 499f45a59..de485bd94 100644
--- a/python/monarch/mesh_controller.py
+++ b/python/monarch/mesh_controller.py
@@ -202,7 +202,6 @@ def _initialize_env(worker_point: Point, proc_id: str) -> None:
         num_worker_procs = len(worker_point.shape)
         process_env = {
             **worker_env,
-            "HYPERACTOR_MANAGED_SUBPROCESS": "1",
             "CUDA_VISIBLE_DEVICES": str(local_rank),
             "NCCL_HOSTID": f"{proc_id}_host_{worker_rank // gpus_per_host}",
             # This is needed to avoid a hard failure in ncclx when we do not
diff --git a/python/monarch/proc_mesh.py b/python/monarch/proc_mesh.py
index 26fb280e8..200ca8cba 100644
--- a/python/monarch/proc_mesh.py
+++ b/python/monarch/proc_mesh.py
@@ -269,7 +269,6 @@ async def proc_mesh_nonblocking(
     env = env or {}
     cmd, args, base_env = _get_bootstrap_args()
     env.update(base_env)
-    env["HYPERACTOR_MANAGED_SUBPROCESS"] = "1"
     allocator = monarch.ProcessAllocator(cmd, args, env)
     alloc = await allocator.allocate(spec)
     return await ProcMesh.from_alloc(alloc)
@@ -284,7 +283,6 @@ def proc_mesh_blocking(
     env = env or {}
     cmd, args, base_env = _get_bootstrap_args()
     env.update(base_env)
-    env["HYPERACTOR_MANAGED_SUBPROCESS"] = "1"
     allocator = monarch.ProcessAllocator(cmd, args, env)
     alloc = allocator.allocate(spec).get()
     return ProcMesh.from_alloc(alloc).get()
diff --git a/python/monarch/rust_local_mesh.py b/python/monarch/rust_local_mesh.py
index 46696b20b..b3fe87b89 100644
--- a/python/monarch/rust_local_mesh.py
+++ b/python/monarch/rust_local_mesh.py
@@ -117,9 +117,7 @@ class ControllerParams(NamedTuple):
     fail_on_worker_timeout: bool
 
 
-_PROC_ENV = {
-    "HYPERACTOR_MANAGED_SUBPROCESS": str(1),
-}
+_PROC_ENV: dict[str, str] = {}
 
 
 def get_controller_main() -> tuple[Path, dict[str, str]]:
@@ -988,7 +986,6 @@ def __init__(
             raise ValueError(f"Unknown socket type: {socket_type}")
 
         env = os.environ.copy()
-        env["HYPERACTOR_MANAGED_SUBPROCESS"] = "1"
         self.env: dict[str, str] = env
 
         # Launch a single system globally
diff --git a/python/monarch/sim_mesh.py b/python/monarch/sim_mesh.py
index dedfc4978..008cc165d 100644
--- a/python/monarch/sim_mesh.py
+++ b/python/monarch/sim_mesh.py
@@ -194,7 +194,6 @@ def __init__(
         fake_call(lambda: 0)
 
         env = os.environ.copy()
-        env["HYPERACTOR_MANAGED_SUBPROCESS"] = "1"
         self.env: dict[str, str] = env
 
         self._mesh_world_state: Dict[MeshWorld, Optional[DeviceMesh]] = mesh_world_state