diff --git a/recipes/audio/bash/native/voice-cloning/.env.example b/recipes/audio/bash/native/voice-cloning/.env.example
index 70a94b2..cdea4b1 100644
--- a/recipes/audio/bash/native/voice-cloning/.env.example
+++ b/recipes/audio/bash/native/voice-cloning/.env.example
@@ -1,2 +1,11 @@
 # Get a key at https://platform.speechify.ai/api-keys
+# Voice cloning must be enabled on your plan.
 SPEECHIFY_API_KEY=
+
+# Full name of the person consenting to have their voice cloned.
+CONSENT_FULL_NAME=Jane Doe
+
+# Optional — override where the recipe reads the audio from.
+# Defaults to sample.wav and consent.wav in the recipe folder.
+# SAMPLE_PATH=./sample.wav
+# CONSENT_RECORDING_PATH=./consent.wav
diff --git a/recipes/audio/bash/native/voice-cloning/README.md b/recipes/audio/bash/native/voice-cloning/README.md
index bd6a429..b6bac54 100644
--- a/recipes/audio/bash/native/voice-cloning/README.md
+++ b/recipes/audio/bash/native/voice-cloning/README.md
@@ -1,7 +1,8 @@
 # Text-to-Speech: voice cloning (Bash, native REST)
 
-Clone a voice from an audio sample, synthesize speech with the clone, then delete it
-— same lifecycle as the [TypeScript](../../../typescript/native/voice-cloning) and
+Clone a voice from an audio sample, synthesize speech with the clone, then delete it, with
+**verified consent** — same lifecycle as the
+[TypeScript](../../../typescript/native/voice-cloning) and
 [Python](../../../python/native/voice-cloning) native recipes, but as a self-contained
 shell script using `curl` + `jq`.
 
@@ -12,6 +13,9 @@ shell script using `curl` + `jq`.
   pointing to [Speechify pricing](https://speechify.ai/pricing) (the API returns
   `402 voice_cloning_not_included`).
 - `bash`, `curl`, `jq`, and `base64`
+- **Two recordings of the same consenting person** (see [Consent](#consent)):
+  - a voice **sample** to clone — 10–30s of clean speech
+  - a **consent recording** — that person reading the challenge phrase this script prints
 
 ## Setup
 
@@ -22,28 +26,43 @@ chmod +x voice-cloning.sh
 
 ## Run
 
+Because consent is verified against a phrase the API generates, this is a two-step run:
+
 ```bash
-./voice-cloning.sh
+./voice-cloning.sh   # 1st run: prints the phrase to read, then exits
+# record the speaker reading that phrase → consent.wav, and their voice → sample.wav
+./voice-cloning.sh   # 2nd run: clones, synthesizes, deletes
 ```
 
-Produces an `output.mp3` spoken in the cloned voice, then removes the cloned voice.
+Put `sample.wav` and `consent.wav` next to the script, or point `SAMPLE_PATH` /
+`CONSENT_RECORDING_PATH` at them. Produces an `output.mp3` in the cloned voice, then
+removes the cloned voice.
 
 ## What it does
 
+- `POST /v1/voices/consent-challenges` (JSON `{ full_name }`, built with `jq -n`) — returns
+  a `phrase` the speaker must read aloud and an `id`; cached in `.consent-challenge.json`
+  between runs (single-use).
 - `POST /v1/voices` (**multipart/form-data**) via `curl -F`: fields `name`, `gender`,
-  `consent`, and a `sample=@fixtures/spacewalk.wav` file part. `curl` sets the
-  `Content-Type` boundary automatically.
-- `POST /v1/audio/speech` with the returned `voice_id`. `jq -n` builds the JSON body
-  safely.
-- `DELETE /v1/voices/{id}` runs from an `EXIT` trap, so the cloned voice is removed
-  even if step 2 fails.
+  `consent_challenge_id`, and `sample=@…` + `consent_recording=@…` file parts. `curl` sets
+  the `Content-Type` boundary automatically.
+- `POST /v1/audio/speech` with the returned `voice_id`.
+- `DELETE /v1/voices/{id}` runs from an `EXIT` trap, so the cloned voice is removed even if
+  a later step fails.
+
+All calls pin `Speechify-Version: 2026-09-13`, the API version the verified-consent flow
+ships on.
 
 ## Consent
 
-`consent` is a **required** JSON string attesting you have the speaker's permission to
-clone their voice (`{"fullName": "...", "email": "..."}`). Only clone voices you are
-authorized to. The bundled `fixtures/spacewalk.wav` is ~26s of NASA ISS spacewalk audio
-(U.S. government work, public domain); replace it with your own consented sample for
-real use.
+Cloning requires **verified consent**. You create a consent challenge, the speaker records
+themselves reading the returned `phrase`, and that recording is sent as `consent_recording`
+and retained as the consent record. The `consent_recording` **must be the same person** as
+the voice `sample` — so there is no shippable sample that comes with valid consent, and this
+recipe is deliberately bring-your-own-audio. Only clone voices you are authorized to.
 
-> Voice cloning reference: https://docs.speechify.ai/tts/guides/voice-cloning
+> The old `consent` JSON field (`fullName` + `email`) is removed on `Speechify-Version:
+> 2026-09-13` — see the
+> [migration guide](https://docs.speechify.ai/build/guides/deprecations/migrating-voice-cloning-consent).
+>
+> Voice cloning reference: https://docs.speechify.ai/build/guides/voice-cloning/overview
diff --git a/recipes/audio/bash/native/voice-cloning/fixtures/spacewalk.wav b/recipes/audio/bash/native/voice-cloning/fixtures/spacewalk.wav
deleted file mode 100644
index 498be74..0000000
Binary files a/recipes/audio/bash/native/voice-cloning/fixtures/spacewalk.wav and /dev/null differ
diff --git a/recipes/audio/bash/native/voice-cloning/voice-cloning.sh b/recipes/audio/bash/native/voice-cloning/voice-cloning.sh
index 78728b3..4f931d5 100755
--- a/recipes/audio/bash/native/voice-cloning/voice-cloning.sh
+++ b/recipes/audio/bash/native/voice-cloning/voice-cloning.sh
@@ -2,7 +2,7 @@
 set -euo pipefail
 
 # Speechify TTS voice cloning (Bash + curl + jq).
-# Full lifecycle: clone → use → delete.
+# Full lifecycle: clone → use → delete, with verified consent.
 
 cd "$(dirname "$0")"
 
@@ -16,24 +16,64 @@ fi
 : "${SPEECHIFY_API_KEY:?Set SPEECHIFY_API_KEY (copy .env.example to .env).}"
 
 BASE="https://api.speechify.ai"
-# Bundled sample: ~26s of NASA ISS spacewalk audio (public domain).
-SAMPLE="fixtures/spacewalk.wav"
-
-# 1. Clone a voice from an audio sample (10-30s of clean speech works well).
-#    POST /v1/voices is multipart/form-data — `curl -F` builds the body and sets
-#    the Content-Type boundary automatically. `consent` is REQUIRED: a JSON
-#    string attesting you have the speaker's permission to clone their voice.
-http_status=0
+# The verified-consent flow ships on this API version — pin it explicitly.
+VERSION="2026-09-13"
+
+# Cloning requires VERIFIED consent: the speaker records themselves reading a
+# phrase the API returns, and that recording is kept as the consent record. It
+# must be the SAME person as the voice sample, so this recipe is
+# bring-your-own-audio — there is no sample that ships with valid consent.
+CONSENT_FULL_NAME="${CONSENT_FULL_NAME:-Jane Doe}"
+SAMPLE="${SAMPLE_PATH:-sample.wav}"
+CONSENT="${CONSENT_RECORDING_PATH:-consent.wav}"
+# The challenge is single-use and its phrase is dynamic, so we cache it between
+# runs: run once to get the phrase, record it, run again to submit.
+CHALLENGE_CACHE=".consent-challenge.json"
+
+auth=(-H "Authorization: Bearer ${SPEECHIFY_API_KEY}" -H "Speechify-Version: ${VERSION}")
+
+# 1. Get (or reuse) a consent challenge. Its `phrase` is what the speaker must
+#    read aloud; `id` ties the recording to this consent on the create.
+if [ ! -f "$CHALLENGE_CACHE" ]; then
+  curl --fail-with-body --silent --show-error \
+    -X POST "${BASE}/v1/voices/consent-challenges" \
+    "${auth[@]}" \
+    -H "Content-Type: application/json" \
+    -d "$(jq -n --arg n "$CONSENT_FULL_NAME" '{full_name: $n}')" \
+    > "$CHALLENGE_CACHE"
+fi
+challenge_id=$(jq -r '.id' < "$CHALLENGE_CACHE")
+phrase=$(jq -r '.phrase' < "$CHALLENGE_CACHE")
+expires_at=$(jq -r '.expires_at' < "$CHALLENGE_CACHE")
+
+# 2. Make sure we have both recordings before spending the (single-use) challenge.
+if [ ! -f "$SAMPLE" ] || [ ! -f "$CONSENT" ]; then
+  echo ""
+  echo "Consent required. Have ${CONSENT_FULL_NAME} record themselves reading this phrase, exactly as written:"
+  echo ""
+  echo "  \"${phrase}\""
+  echo ""
+  echo "Then provide these files and re-run (paths override via SAMPLE_PATH / CONSENT_RECORDING_PATH):"
+  [ ! -f "$SAMPLE" ]  && echo "  sample:  ${SAMPLE}  (${CONSENT_FULL_NAME}'s voice, 10-30s of clean speech)"
+  [ ! -f "$CONSENT" ] && echo "  consent: ${CONSENT}  (the SAME person reading the phrase above)"
+  echo ""
+  echo "Challenge expires ${expires_at}. If it has expired, delete ${CHALLENGE_CACHE} and re-run."
+  exit 1
+fi
+
+# 3. Clone the voice. POST /v1/voices is multipart/form-data — `curl -F` builds
+#    the body and sets the Content-Type boundary automatically.
 create_body=$(mktemp)
 trap 'rm -f "$create_body"' EXIT
 
 http_status=$(curl --silent --show-error --output "$create_body" --write-out '%{http_code}' \
   -X POST "${BASE}/v1/voices" \
-  -H "Authorization: Bearer ${SPEECHIFY_API_KEY}" \
+  "${auth[@]}" \
   -F "name=cookbook-cloned-voice" \
   -F "gender=male" \
-  -F 'consent={"fullName":"Jane Doe","email":"jane@example.com"};type=text/plain' \
-  -F "sample=@${SAMPLE};type=audio/wav")
+  -F "consent_challenge_id=${challenge_id}" \
+  -F "sample=@${SAMPLE}" \
+  -F "consent_recording=@${CONSENT}")
 
 if [ "$http_status" = "402" ]; then
   echo ""
@@ -45,19 +85,22 @@ if [ "$http_status" -lt 200 ] || [ "$http_status" -ge 300 ]; then
   echo "POST /v1/voices → ${http_status}" >&2
   cat "$create_body" >&2
   echo >&2
+  echo "If this is a consent_* error, the challenge may be spent or expired — delete ${CHALLENGE_CACHE} and re-run." >&2
   exit 1
 fi
 
+rm -f "$CHALLENGE_CACHE"  # challenge is spent
+
 voice_id=$(jq -r '.id' < "$create_body")
 display_name=$(jq -r '.display_name' < "$create_body")
 voice_type=$(jq -r '.type' < "$create_body")
 echo "Cloned voice created: ${voice_id} (${display_name}, type=${voice_type})"
 
-# Ensure we always delete the cloned voice, even on failure of step 2.
+# Ensure we always delete the cloned voice, even on failure of step 4.
 cleanup() {
   del_status=$(curl --silent --show-error --output /dev/null --write-out '%{http_code}' \
     -X DELETE "${BASE}/v1/voices/${voice_id}" \
-    -H "Authorization: Bearer ${SPEECHIFY_API_KEY}")
+    "${auth[@]}")
   if [ "$del_status" -ge 200 ] && [ "$del_status" -lt 300 ]; then
     echo "Deleted cloned voice ${voice_id}"
   else
@@ -66,10 +109,10 @@ cleanup() {
 }
 trap 'cleanup; rm -f "$create_body"' EXIT
 
-# 2. Synthesize speech using the cloned voice — pass its id as voice_id.
+# 4. Synthesize speech using the cloned voice — pass its id as voice_id.
 speech_response=$(curl --fail-with-body --silent --show-error \
   -X POST "${BASE}/v1/audio/speech" \
-  -H "Authorization: Bearer ${SPEECHIFY_API_KEY}" \
+  "${auth[@]}" \
   -H "Content-Type: application/json" \
   -d "$(jq -n --arg vid "$voice_id" '{
     input: "Hello from a voice cloned with the Speechify API.",
@@ -81,4 +124,4 @@ speech_response=$(curl --fail-with-body --silent --show-error \
 printf '%s' "$speech_response" | jq -r '.audio_data' | base64 -d > output.mp3
 echo "Wrote output.mp3"
 
-# 3. Cleanup runs from the EXIT trap.
+# 5. Cleanup runs from the EXIT trap.
diff --git a/recipes/audio/python/native/voice-cloning/.env.example b/recipes/audio/python/native/voice-cloning/.env.example
index 4cb51e4..cdea4b1 100644
--- a/recipes/audio/python/native/voice-cloning/.env.example
+++ b/recipes/audio/python/native/voice-cloning/.env.example
@@ -1 +1,11 @@
+# Get a key at https://platform.speechify.ai/api-keys
+# Voice cloning must be enabled on your plan.
 SPEECHIFY_API_KEY=
+
+# Full name of the person consenting to have their voice cloned.
+CONSENT_FULL_NAME=Jane Doe
+
+# Optional — override where the recipe reads the audio from.
+# Defaults to sample.wav and consent.wav in the recipe folder.
+# SAMPLE_PATH=./sample.wav
+# CONSENT_RECORDING_PATH=./consent.wav
diff --git a/recipes/audio/python/native/voice-cloning/README.md b/recipes/audio/python/native/voice-cloning/README.md
index 862c7dd..493f00b 100644
--- a/recipes/audio/python/native/voice-cloning/README.md
+++ b/recipes/audio/python/native/voice-cloning/README.md
@@ -1,8 +1,8 @@
 # Text-to-Speech: voice cloning (Python, native REST)
 
-The same as [`voice-cloning`](../../sdk/voice-cloning) — clone → synthesize → delete —
-but calling the REST API directly with `requests` + multipart form-data instead of the
-`speechify-api` SDK.
+The same as [`voice-cloning`](../../sdk/voice-cloning) — clone → synthesize → delete, with
+**verified consent** — but calling the REST API directly with `requests` + multipart
+form-data instead of the `speechify-api` SDK.
 
 ## Prerequisites
 
@@ -11,6 +11,9 @@ but calling the REST API directly with `requests` + multipart form-data instead
   pointing to [Speechify pricing](https://speechify.ai/pricing) (the API returns
   `402 voice_cloning_not_included`).
 - Python 3.10+ and [uv](https://docs.astral.sh/uv/)
+- **Two recordings of the same consenting person** (see [Consent](#consent)):
+  - a voice **sample** to clone — 10–30s of clean speech
+  - a **consent recording** — that person reading the challenge phrase this recipe prints
 
 ## Setup
 
@@ -21,26 +24,43 @@ uv sync
 
 ## Run
 
+Because consent is verified against a phrase the API generates, this is a two-step run:
+
 ```bash
-uv run main.py
+uv run main.py   # 1st run: prints the phrase to read, then exits
+# record the speaker reading that phrase → consent.wav, and their voice → sample.wav
+uv run main.py   # 2nd run: clones, synthesizes, deletes
 ```
 
-Produces an `output.mp3` spoken in the cloned voice, then removes the cloned voice.
+Put `sample.wav` and `consent.wav` in the recipe folder, or point `SAMPLE_PATH` /
+`CONSENT_RECORDING_PATH` at them. Produces an `output.mp3` in the cloned voice, then
+removes the cloned voice.
 
 ## What it does
 
-- `POST /v1/voices` (**multipart/form-data**): fields `name`, `gender`, `consent`, and
-  a `sample` file part. Pass them via `data=` + `files=` and `requests` sets the
-  `Content-Type` boundary automatically — do **not** set it yourself.
+- `POST /v1/voices/consent-challenges` (JSON `{ full_name }`) — returns a `phrase` the
+  speaker must read aloud and an `id`; cached in `.consent-challenge.json` between runs
+  (single-use).
+- `POST /v1/voices` (**multipart/form-data**): fields `name`, `gender`,
+  `consent_challenge_id`, and `sample` + `consent_recording` file parts. Pass them via
+  `data=` + `files=` and `requests` sets the `Content-Type` boundary — do **not** set it
+  yourself.
 - `POST /v1/audio/speech` with the returned voice's `id` as `voice_id`.
 - `DELETE /v1/voices/{id}` — cleans up so personal voices don't accumulate.
 
+All calls pin `Speechify-Version: 2026-09-13`, the API version the verified-consent flow
+ships on.
+
 ## Consent
 
-`consent` is a **required** JSON string attesting you have the speaker's permission to
-clone their voice (`{"fullName": "...", "email": "..."}`). Only clone voices you are
-authorized to. The bundled `fixtures/spacewalk.wav` is ~26s of NASA ISS spacewalk audio
-(U.S. government work, public domain); replace it with your own consented sample for
-real use.
+Cloning requires **verified consent**. You create a consent challenge, the speaker records
+themselves reading the returned `phrase`, and that recording is sent as `consent_recording`
+and retained as the consent record. The `consent_recording` **must be the same person** as
+the voice `sample` — so there is no shippable sample that comes with valid consent, and this
+recipe is deliberately bring-your-own-audio. Only clone voices you are authorized to.
 
-> Voice cloning reference: https://docs.speechify.ai/tts/guides/voice-cloning
+> The old `consent` JSON field (`fullName` + `email`) is removed on `Speechify-Version:
+> 2026-09-13` — see the
+> [migration guide](https://docs.speechify.ai/build/guides/deprecations/migrating-voice-cloning-consent).
+>
+> Voice cloning reference: https://docs.speechify.ai/build/guides/voice-cloning/overview
diff --git a/recipes/audio/python/native/voice-cloning/fixtures/spacewalk.wav b/recipes/audio/python/native/voice-cloning/fixtures/spacewalk.wav
deleted file mode 100644
index 498be74..0000000
Binary files a/recipes/audio/python/native/voice-cloning/fixtures/spacewalk.wav and /dev/null differ
diff --git a/recipes/audio/python/native/voice-cloning/main.py b/recipes/audio/python/native/voice-cloning/main.py
index 201cd70..19b4163 100644
--- a/recipes/audio/python/native/voice-cloning/main.py
+++ b/recipes/audio/python/native/voice-cloning/main.py
@@ -1,6 +1,7 @@
 import base64
 import json
 import os
+from datetime import datetime, timezone
 
 import requests
 from dotenv import load_dotenv
@@ -10,9 +11,31 @@
 # form-data instead of the speechify-api SDK.
 
 BASE = "https://api.speechify.ai"
+# The verified-consent flow ships on this API version — pin it explicitly.
+VERSION = "2026-09-13"
 
-# Bundled sample: ~26s of NASA ISS spacewalk audio (public domain).
-SAMPLE_PATH = os.path.join(os.path.dirname(__file__), "fixtures", "spacewalk.wav")
+# Cloning requires VERIFIED consent: the speaker records themselves reading a
+# phrase the API returns, and that recording is kept as the consent record. It
+# must be the SAME person as the voice sample, so this recipe is
+# bring-your-own-audio — there is no sample that ships with valid consent.
+HERE = os.path.dirname(__file__)
+CONSENT_FULL_NAME = os.environ.get("CONSENT_FULL_NAME", "Jane Doe")
+SAMPLE_PATH = os.environ.get("SAMPLE_PATH", os.path.join(HERE, "sample.wav"))
+CONSENT_RECORDING_PATH = os.environ.get("CONSENT_RECORDING_PATH", os.path.join(HERE, "consent.wav"))
+# The challenge is single-use and its phrase is dynamic, so we cache it between
+# runs: run once to get the phrase, record it, run again to submit.
+CHALLENGE_CACHE = os.path.join(HERE, ".consent-challenge.json")
+
+
+def load_challenge():
+    if not os.path.exists(CHALLENGE_CACHE):
+        return None
+    with open(CHALLENGE_CACHE) as f:
+        c = json.load(f)
+    # Python 3.10's fromisoformat doesn't accept a trailing "Z".
+    if datetime.fromisoformat(c["expires_at"].replace("Z", "+00:00")) <= datetime.now(timezone.utc):
+        return None  # expired -> make a fresh one
+    return c
 
 
 def main() -> None:
@@ -22,22 +45,54 @@ def main() -> None:
     if not token:
         raise SystemExit("Set SPEECHIFY_API_KEY (copy .env.example to .env).")
 
-    auth = {"Authorization": f"Bearer {token}"}
+    auth = {"Authorization": f"Bearer {token}", "Speechify-Version": VERSION}
+
+    # 1. Get (or reuse) a consent challenge. Its `phrase` is what the speaker must
+    #    read aloud; `id` ties the recording to this consent on the create.
+    challenge = load_challenge()
+    if challenge is None:
+        resp = requests.post(
+            f"{BASE}/v1/voices/consent-challenges",
+            headers={**auth, "Content-Type": "application/json"},
+            json={"full_name": CONSENT_FULL_NAME},
+            timeout=30,
+        )
+        resp.raise_for_status()
+        challenge = resp.json()
+        with open(CHALLENGE_CACHE, "w") as f:
+            json.dump(challenge, f, indent=2)
 
-    # 1. Clone a voice from an audio sample (10-30s of clean speech works well).
-    #    POST /v1/voices is multipart/form-data — pass `files=` to requests and it
-    #    sets the Content-Type boundary automatically. `consent` is REQUIRED: a JSON
-    #    string attesting you have the speaker's permission to clone their voice.
-    with open(SAMPLE_PATH, "rb") as sample:
+    # 2. Make sure we have both recordings before spending the (single-use) challenge.
+    missing = []
+    if not os.path.exists(SAMPLE_PATH):
+        missing.append(f"  sample:  {SAMPLE_PATH}  ({CONSENT_FULL_NAME}'s voice, 10-30s of clean speech)")
+    if not os.path.exists(CONSENT_RECORDING_PATH):
+        missing.append(f"  consent: {CONSENT_RECORDING_PATH}  (the SAME person reading the phrase below)")
+    if missing:
+        raise SystemExit(
+            f"\nConsent required. Have {CONSENT_FULL_NAME} record themselves reading this phrase, "
+            "exactly as written:\n\n"
+            f'  "{challenge["phrase"]}"\n\n'
+            "Then provide these files and re-run (paths override via SAMPLE_PATH / CONSENT_RECORDING_PATH):\n"
+            + "\n".join(missing)
+            + f"\n\nChallenge expires {challenge['expires_at']}.\n"
+        )
+
+    # 3. Clone the voice. POST /v1/voices is multipart/form-data — pass `files=` to
+    #    requests and it sets the Content-Type boundary automatically.
+    with open(SAMPLE_PATH, "rb") as sample, open(CONSENT_RECORDING_PATH, "rb") as consent:
         create_resp = requests.post(
             f"{BASE}/v1/voices",
             headers=auth,
             data={
                 "name": "cookbook-cloned-voice",
                 "gender": "male",
-                "consent": json.dumps({"fullName": "Jane Doe", "email": "jane@example.com"}),
+                "consent_challenge_id": challenge["id"],
+            },
+            files={
+                "sample": (os.path.basename(SAMPLE_PATH), sample),
+                "consent_recording": (os.path.basename(CONSENT_RECORDING_PATH), consent),
             },
-            files={"sample": ("spacewalk.wav", sample, "audio/wav")},
             timeout=120,
         )
 
@@ -47,11 +102,12 @@ def main() -> None:
             "Upgrade to a plan that includes voice cloning: https://speechify.ai/pricing\n"
         )
     create_resp.raise_for_status()
+    os.remove(CHALLENGE_CACHE)  # challenge is spent
     voice = create_resp.json()
     print(f"Cloned voice created: {voice['id']} ({voice['display_name']}, type={voice['type']})")
 
     try:
-        # 2. Synthesize speech using the cloned voice — pass its id as voice_id.
+        # 4. Synthesize speech using the cloned voice — pass its id as voice_id.
         speech_resp = requests.post(
             f"{BASE}/v1/audio/speech",
             headers={**auth, "Content-Type": "application/json"},
@@ -69,7 +125,7 @@ def main() -> None:
             f.write(base64.b64decode(speech["audio_data"]))
         print("Wrote output.mp3")
     finally:
-        # 3. Clean up so cloned voices don't accumulate on your account.
+        # 5. Clean up so cloned voices don't accumulate on your account.
         #    Remove this to keep the voice and reuse it later via voice.id.
         del_resp = requests.delete(f"{BASE}/v1/voices/{voice['id']}", headers=auth, timeout=30)
         if del_resp.ok:
diff --git a/recipes/audio/typescript/native/voice-cloning/.env.example b/recipes/audio/typescript/native/voice-cloning/.env.example
index 70a94b2..cdea4b1 100644
--- a/recipes/audio/typescript/native/voice-cloning/.env.example
+++ b/recipes/audio/typescript/native/voice-cloning/.env.example
@@ -1,2 +1,11 @@
 # Get a key at https://platform.speechify.ai/api-keys
+# Voice cloning must be enabled on your plan.
 SPEECHIFY_API_KEY=
+
+# Full name of the person consenting to have their voice cloned.
+CONSENT_FULL_NAME=Jane Doe
+
+# Optional — override where the recipe reads the audio from.
+# Defaults to sample.wav and consent.wav in the recipe folder.
+# SAMPLE_PATH=./sample.wav
+# CONSENT_RECORDING_PATH=./consent.wav
diff --git a/recipes/audio/typescript/native/voice-cloning/README.md b/recipes/audio/typescript/native/voice-cloning/README.md
index 02bb8fe..f46ce1f 100644
--- a/recipes/audio/typescript/native/voice-cloning/README.md
+++ b/recipes/audio/typescript/native/voice-cloning/README.md
@@ -1,8 +1,8 @@
 # Text-to-Speech: voice cloning (TypeScript, native REST)
 
-The same as [`voice-cloning`](../../sdk/voice-cloning) — clone → synthesize → delete —
-but calling the REST API directly with `fetch` + multipart `FormData` instead of the
-`@speechify/api` SDK.
+The same as [`voice-cloning`](../../sdk/voice-cloning) — clone → synthesize → delete, with
+**verified consent** — but calling the REST API directly with `fetch` + multipart
+`FormData` instead of the `@speechify/api` SDK.
 
 ## Prerequisites
 
@@ -11,6 +11,9 @@ but calling the REST API directly with `fetch` + multipart `FormData` instead of
   pointing to [Speechify pricing](https://speechify.ai/pricing) (the API returns
   `402 voice_cloning_not_included`).
 - Node 20+ (for built-in `fetch`, `FormData`, and `Blob`)
+- **Two recordings of the same consenting person** (see [Consent](#consent)):
+  - a voice **sample** to clone — 10–30s of clean speech
+  - a **consent recording** — that person reading the challenge phrase this recipe prints
 
 ## Setup
 
@@ -21,26 +24,43 @@ pnpm install
 
 ## Run
 
+Because consent is verified against a phrase the API generates, this is a two-step run:
+
 ```bash
-pnpm start
+pnpm start   # 1st run: prints the phrase to read, then exits
+# record the speaker reading that phrase → consent.wav, and their voice → sample.wav
+pnpm start   # 2nd run: clones, synthesizes, deletes
 ```
 
-Produces an `output.mp3` spoken in the cloned voice, then removes the cloned voice.
+Put `sample.wav` and `consent.wav` in the recipe folder, or point `SAMPLE_PATH` /
+`CONSENT_RECORDING_PATH` at them. Produces an `output.mp3` in the cloned voice, then
+removes the cloned voice.
 
 ## What it does
 
-- `POST /v1/voices` (**multipart/form-data**): fields `name`, `gender`, `consent`, and a
-  `sample` file part. Pass a `FormData` instance as `body` and let `fetch` set the
-  `Content-Type` boundary automatically — do **not** set it yourself.
+- `POST /v1/voices/consent-challenges` (JSON `{ full_name }`) — returns a `phrase` the
+  speaker must read aloud and an `id`; cached in `.consent-challenge.json` between runs
+  (single-use).
+- `POST /v1/voices` (**multipart/form-data**): fields `name`, `gender`,
+  `consent_challenge_id`, and `sample` + `consent_recording` file parts. Pass a
+  `FormData` instance as `body` and let `fetch` set the `Content-Type` boundary — do
+  **not** set it yourself.
 - `POST /v1/audio/speech` with the returned voice's `id` as `voice_id`.
 - `DELETE /v1/voices/{id}` — cleans up so personal voices don't accumulate.
 
+All calls pin `Speechify-Version: 2026-09-13`, the API version the verified-consent flow
+ships on.
+
 ## Consent
 
-`consent` is a **required** JSON string attesting you have the speaker's permission to
-clone their voice (`{"fullName": "...", "email": "..."}`). Only clone voices you are
-authorized to. The bundled `fixtures/spacewalk.wav` is ~26s of NASA ISS spacewalk audio
-(U.S. government work, public domain); replace it with your own consented sample for
-real use.
+Cloning requires **verified consent**. You create a consent challenge, the speaker records
+themselves reading the returned `phrase`, and that recording is sent as `consent_recording`
+and retained as the consent record. The `consent_recording` **must be the same person** as
+the voice `sample` — so there is no shippable sample that comes with valid consent, and this
+recipe is deliberately bring-your-own-audio. Only clone voices you are authorized to.
 
-> Voice cloning reference: https://docs.speechify.ai/tts/guides/voice-cloning
+> The old `consent` JSON field (`fullName` + `email`) is removed on `Speechify-Version:
+> 2026-09-13` — see the
+> [migration guide](https://docs.speechify.ai/build/guides/deprecations/migrating-voice-cloning-consent).
+>
+> Voice cloning reference: https://docs.speechify.ai/build/guides/voice-cloning/overview
diff --git a/recipes/audio/typescript/native/voice-cloning/fixtures/spacewalk.wav b/recipes/audio/typescript/native/voice-cloning/fixtures/spacewalk.wav
deleted file mode 100644
index 498be74..0000000
Binary files a/recipes/audio/typescript/native/voice-cloning/fixtures/spacewalk.wav and /dev/null differ
diff --git a/recipes/audio/typescript/native/voice-cloning/src/index.ts b/recipes/audio/typescript/native/voice-cloning/src/index.ts
index 8513250..91e7f02 100644
--- a/recipes/audio/typescript/native/voice-cloning/src/index.ts
+++ b/recipes/audio/typescript/native/voice-cloning/src/index.ts
@@ -12,10 +12,27 @@ if (!token) {
 }
 
 const BASE = "https://api.speechify.ai";
+// The verified-consent flow ships on this API version — pin it explicitly.
+const VERSION = "2026-09-13";
+const authHeaders = { Authorization: `Bearer ${token}`, "Speechify-Version": VERSION };
 
-// Bundled sample: ~26s of NASA ISS spacewalk audio (public domain).
-const samplePath = path.resolve(import.meta.dirname, "../fixtures/spacewalk.wav");
+// Cloning requires VERIFIED consent: the speaker records themselves reading a
+// phrase the API returns, and that recording is kept as the consent record. It
+// must be the SAME person as the voice sample, so this recipe is
+// bring-your-own-audio — there is no sample that ships with valid consent.
+const dir = import.meta.dirname;
+const CONSENT_FULL_NAME = process.env.CONSENT_FULL_NAME ?? "Jane Doe";
+const samplePath = path.resolve(process.env.SAMPLE_PATH ?? path.join(dir, "../sample.wav"));
+const consentPath = path.resolve(process.env.CONSENT_RECORDING_PATH ?? path.join(dir, "../consent.wav"));
+// The challenge is single-use and its phrase is dynamic, so we cache it between
+// runs: run once to get the phrase, record it, run again to submit.
+const challengeCache = path.join(dir, "../.consent-challenge.json");
 
+interface Challenge {
+  id: string;
+  phrase: string;
+  expires_at: string;
+}
 interface CreatedVoice {
   id: string;
   display_name: string;
@@ -27,23 +44,58 @@ interface SpeechResponse {
   billable_characters_count: number;
 }
 
+function loadChallenge(): Challenge | null {
+  if (!fs.existsSync(challengeCache)) return null;
+  const c = JSON.parse(fs.readFileSync(challengeCache, "utf8")) as Challenge;
+  if (new Date(c.expires_at).getTime() <= Date.now()) return null; // expired → make a fresh one
+  return c;
+}
+
 async function main() {
-  // 1. Clone a voice from an audio sample (10–30s of clean speech works well).
-  //    POST /v1/voices is multipart/form-data — let fetch set the boundary by
-  //    passing a FormData instance directly (do NOT set Content-Type manually).
-  //    `consent` is REQUIRED: a JSON string attesting you have the speaker's
-  //    permission to clone their voice.
+  // 1. Get (or reuse) a consent challenge. Its `phrase` is what the speaker must
+  //    read aloud; `id` ties the recording to this consent on the create.
+  let challenge = loadChallenge();
+  if (!challenge) {
+    const res = await fetch(`${BASE}/v1/voices/consent-challenges`, {
+      method: "POST",
+      headers: { ...authHeaders, "Content-Type": "application/json" },
+      body: JSON.stringify({ full_name: CONSENT_FULL_NAME }),
+    });
+    if (!res.ok) {
+      throw new Error(`POST /v1/voices/consent-challenges → ${res.status} ${res.statusText}: ${await res.text()}`);
+    }
+    challenge = (await res.json()) as Challenge;
+    fs.writeFileSync(challengeCache, JSON.stringify(challenge, null, 2));
+  }
+
+  // 2. Make sure we have both recordings before spending the (single-use) challenge.
+  const missing = [
+    fs.existsSync(samplePath) ? null : `  sample:  ${samplePath}  (${CONSENT_FULL_NAME}'s voice, 10–30s of clean speech)`,
+    fs.existsSync(consentPath) ? null : `  consent: ${consentPath}  (the SAME person reading the phrase below)`,
+  ].filter(Boolean);
+  if (missing.length > 0) {
+    console.log(
+      `\nConsent required. Have ${CONSENT_FULL_NAME} record themselves reading this phrase, exactly as written:\n\n` +
+        `  "${challenge.phrase}"\n\n` +
+        `Then provide these files and re-run (paths override via SAMPLE_PATH / CONSENT_RECORDING_PATH):\n` +
+        missing.join("\n") +
+        `\n\nChallenge expires ${challenge.expires_at}.\n`,
+    );
+    process.exit(1);
+  }
+
+  // 3. Clone the voice. POST /v1/voices is multipart/form-data — let fetch set the
+  //    boundary by passing a FormData instance directly (do NOT set Content-Type).
   const form = new FormData();
   form.append("name", "cookbook-cloned-voice");
   form.append("gender", "male");
-  form.append("consent", JSON.stringify({ fullName: "Jane Doe", email: "jane@example.com" }));
-  // Wrap the file bytes in a Blob with the original filename for the multipart part.
-  const sampleBytes = fs.readFileSync(samplePath);
-  form.append("sample", new Blob([sampleBytes], { type: "audio/wav" }), "spacewalk.wav");
+  form.append("consent_challenge_id", challenge.id);
+  form.append("sample", new Blob([fs.readFileSync(samplePath)]), path.basename(samplePath));
+  form.append("consent_recording", new Blob([fs.readFileSync(consentPath)]), path.basename(consentPath));
 
   const createRes = await fetch(`${BASE}/v1/voices`, {
     method: "POST",
-    headers: { Authorization: `Bearer ${token}` },
+    headers: authHeaders,
     body: form,
   });
 
@@ -60,18 +112,16 @@ async function main() {
       `POST /v1/voices → ${createRes.status} ${createRes.statusText}: ${await createRes.text()}`,
     );
   }
+  fs.rmSync(challengeCache, { force: true }); // challenge is spent
 
   const voice = (await createRes.json()) as CreatedVoice;
   console.log(`Cloned voice created: ${voice.id} (${voice.display_name}, type=${voice.type})`);
 
   try {
-    // 2. Synthesize speech using the cloned voice — pass its id as voice_id.
+    // 4. Synthesize speech using the cloned voice — pass its id as voice_id.
     const speechRes = await fetch(`${BASE}/v1/audio/speech`, {
       method: "POST",
-      headers: {
-        Authorization: `Bearer ${token}`,
-        "Content-Type": "application/json",
-      },
+      headers: { ...authHeaders, "Content-Type": "application/json" },
       body: JSON.stringify({
         input: "Hello from a voice cloned with the Speechify API.",
         voice_id: voice.id,
@@ -88,11 +138,11 @@ async function main() {
     fs.writeFileSync("output.mp3", Buffer.from(speech.audio_data, "base64"));
     console.log("Wrote output.mp3");
   } finally {
-    // 3. Clean up so cloned voices don't accumulate on your account.
+    // 5. Clean up so cloned voices don't accumulate on your account.
     //    Remove this to keep the voice and reuse it later via voice.id.
     const delRes = await fetch(`${BASE}/v1/voices/${encodeURIComponent(voice.id)}`, {
       method: "DELETE",
-      headers: { Authorization: `Bearer ${token}` },
+      headers: authHeaders,
     });
     if (!delRes.ok) {
       console.error(