diff --git a/src/openhuman/voice/cloud_transcribe.rs b/src/openhuman/voice/cloud_transcribe.rs
index a88591c26..f7f121d5c 100644
--- a/src/openhuman/voice/cloud_transcribe.rs
+++ b/src/openhuman/voice/cloud_transcribe.rs
@@ -22,6 +22,10 @@ use crate::rpc::RpcOutcome;
 
 const LOG_PREFIX: &str = "[voice_cloud_stt]";
 
+/// Maximum base64 audio input length (~25MB base64 → ~18MB decoded audio).
+/// Prevents OOM on unreasonably large payloads (multi-hour recordings).
+const MAX_AUDIO_BASE64_LEN: usize = 33_554_432;
+
 /// Default model id sent to the backend. The backend's controller currently
 /// resolves this to whichever provider it has configured for audio
 /// transcription (today: GMI Whisper). Callers can override.
@@ -56,6 +60,12 @@ pub async fn transcribe_cloud(
     if trimmed.is_empty() {
         return Err("audio_base64 is required".to_string());
     }
+    if trimmed.len() > MAX_AUDIO_BASE64_LEN {
+        return Err(format!(
+            "audio_base64 exceeds maximum size ({}MB)",
+            MAX_AUDIO_BASE64_LEN / 1_048_576
+        ));
+    }
     let audio_bytes = BASE64
         .decode(trimmed)
         .map_err(|e| format!("invalid base64 audio: {e}"))?;