diff --git a/ddtrace/internal/core/crashtracking.py b/ddtrace/internal/core/crashtracking.py
index 253ab133bc9..02505672ec8 100644
--- a/ddtrace/internal/core/crashtracking.py
+++ b/ddtrace/internal/core/crashtracking.py
@@ -19,7 +19,6 @@
 from ddtrace.internal.settings.crashtracker import config as crashtracker_config
 from ddtrace.internal.settings.profiling import config as profiling_config
 from ddtrace.internal.settings.profiling import config_str
-from ddtrace.internal.telemetry.writer import _agentless_endpoint_url
 
 
 log = get_logger(__name__)
@@ -125,17 +124,12 @@ def _get_args(additional_tags: Optional[dict[str, str]]):
         log.error("Invalid stacktrace_resolver value: %s", crashtracker_config.stacktrace_resolver)
         stacktrace_resolver = StacktraceCollection.EnabledWithInprocessSymbols
 
-    # Crash reports ride the telemetry intake, so agentless points at the same host the telemetry
-    # writer uses. libdatadog only resolves the direct intake path (rather than the agent's
-    # telemetry proxy path) when the endpoint carries an API key and direct submission is
-    # enabled in the receiver process, so both are set together below.
+    # Do not manually compute an url. The crashtracker receiver has this handling,
+    # given DD_API_KEY and DD_SITE. Nothing to do for us here.
+    # We even cannot do so, otherwise we'll pin crashtracking to a single host,
+    # instead of dedicated error-reporting and crashtracking hosts.
     crash_agentless = config._agentless_enabled and config._dd_api_key
-    if crash_agentless:
-        upload_url = _agentless_endpoint_url(config._dd_site)
-        api_key = config._dd_api_key
-    else:
-        upload_url = agent_config.trace_agent_url
-        api_key = None
+    upload_url = None if crash_agentless else agent_config.trace_agent_url
 
     # Create crashtracker configuration
     crashtracker_configuration = CrashtrackerConfiguration(
@@ -149,13 +143,15 @@ def _get_args(additional_tags: Optional[dict[str, str]]):
         crashtracker_config.debug_url or upload_url,
         None,  # unix_socket_path
         crashtracker_config._test_token,
-        api_key,
+        None,
     )
 
     receiver_env = {}
 
     if crash_agentless:
         receiver_env["_DD_DIRECT_SUBMISSION_ENABLED"] = "true"
+        receiver_env["DD_API_KEY"] = config._dd_api_key
+        receiver_env["DD_SITE"] = config._dd_site
 
     # Don't pass all env vars to the receiver process, because there are
     # conflicts with export location derivation
diff --git a/releasenotes/notes/agentless-crashtracking-fix-endpoint-e3e4b73aad92c90f.yaml b/releasenotes/notes/agentless-crashtracking-fix-endpoint-e3e4b73aad92c90f.yaml
new file mode 100644
index 00000000000..71891bb2791
--- /dev/null
+++ b/releasenotes/notes/agentless-crashtracking-fix-endpoint-e3e4b73aad92c90f.yaml
@@ -0,0 +1,7 @@
+---
+fixes:
+  - |
+    crashtracking: fix crash reports being sent to the wrong intake when ``DD_AGENTLESS_ENABLED``
+    is set. The errors-intake upload was incorrectly reusing the same endpoint configured for the
+    telemetry crash-report path, sending it to the telemetry intake host instead of the dedicated
+    errors-intake host.
\ No newline at end of file
diff --git a/tests/crashtracker/test_crashtracker.py b/tests/crashtracker/test_crashtracker.py
index 4ba215fcd65..bc8d211729f 100644
--- a/tests/crashtracker/test_crashtracker.py
+++ b/tests/crashtracker/test_crashtracker.py
@@ -1073,10 +1073,13 @@ def test_crashtracker_unhandled_exception(run_python_code_in_subprocess):
 
 @pytest.mark.subprocess(env={"DD_AGENTLESS_ENABLED": "true", "DD_API_KEY": "foobarkey"})
 def test_crashtracker_uploads_to_the_intake_when_agentless():
-    """Crash reports ride the telemetry intake, so agentless has to supply a key with the URL.
+    """Agentless must not pin crashtracking to a single, explicitly-computed endpoint.
 
-    libdatadog only resolves the direct intake path over the agent's telemetry proxy path when the
-    endpoint carries an API key *and* the receiver has direct submission enabled.
+    The crash-report (telemetry intake) and errors-intake uploads go to genuinely different
+    hosts. Forcing one endpoint on both would send the errors-intake upload to the wrong host,
+    so we pass no endpoint/api_key at all here - the receiver resolves both independently from
+    DD_API_KEY/DD_SITE/_DD_DIRECT_SUBMISSION_ENABLED, which must be forwarded to it explicitly
+    since its environment is not inherited (it's spawned via a raw execve).
     """
     from unittest import mock
 
@@ -1099,9 +1102,11 @@ def fake_receiver(args, env, *rest):
     ):
         crashtracking._get_args({})
 
-    assert captured["endpoint"] == "https://instrumentation-telemetry-intake.datadoghq.com/"
-    assert captured["api_key"] == "foobarkey"
+    assert captured["endpoint"] is None
+    assert captured["api_key"] is None
     assert captured["receiver_env"]["_DD_DIRECT_SUBMISSION_ENABLED"] == "true"
+    assert captured["receiver_env"]["DD_API_KEY"] == "foobarkey"
+    assert captured["receiver_env"]["DD_SITE"] == "datadoghq.com"
 
 
 @pytest.mark.subprocess(env={"DD_API_KEY": "foobarkey"})