GPAI Ledger › GPAI Training Transparency tracker (AI Accountability Lab (AIAL)) › Capture 31 Aug 2026

ElevenLabs_2026_08_03 — capture 20260831T103027Z

Filed under AI Accountability Lab (AIAL) — GPAI Training Transparency tracker, the source this project captured it from; the assessment itself is of the model named above.

Verify: sha256sum c6e2947ba8335be352a9b3009c7359520326e241e8db872f65c7d73ac0ebadf7.pdf must equal the hash above (the filename IS the expected hash); ots verify c6e2947ba8335be352a9b3009c7359520326e241e8db872f65c7d73ac0ebadf7.pdf.20260831T103027Z.ots -f c6e2947ba8335be352a9b3009c7359520326e241e8db872f65c7d73ac0ebadf7.pdf (opentimestamps.org) proves the bytes existed no later than the attestation time — an upper bound on the capture time; the fetch time above is the archive's own record (fresh proofs report 'pending' until bitcoin-anchored, typically within a day). ots verify needs a local Bitcoin Core node (a pruned one is fine); without one, ots info on the proof prints the attesting block height and merkle path to check on any block explorer.

Extracted text

Machine-extracted text (layout may be lost; the authoritative content is the stored file above).

Training  Data  Transparency  Disclosure
 Developer:  Eleven  Labs  Inc.  (“ ElevenLabs ”)  Last  Updated:  January  1,  2026  Effective  Date:  January  1,  2026  Jurisdictional  Applicability:  California  (AB  2013)  and  other  jurisdictions  where  training  data
transparency

disclosures

are

required

or

otherwise

provided.
  This  disclosure  (“ Disclosure ”)  is  provided  to  promote  transparency  regarding  the  data  used  to
train

ElevenLabs’

generative

artificial

intelligence

(“
AI
”)

systems

and

services.

It

is

intended

to

satisfy

applicable

training

data

transparency

and

disclosure

requirements,

including

California

Civil

Code

§§

3110-3111

(AB

2013),

and

to

provide

similar

transparency

information

to

users

in

other

jurisdictions

where

comparable

transparency

obligations

apply.

Model  Families  Covered  by  This  Disclosure
ElevenLabs  organizes  its  generative  AI  systems  into  model  families  based  on  modality,  intended
function,

and

training

data

characteristics.

Each

model

family

described

below

constitutes

a

distinct

“generative

AI

system

or

service”

for

purposes

of

AB

2013

and

analogous

transparency

frameworks.

1.  Text-to-Speech  (TTS)  Models
Operational  domain :  AI  audio  –  synthetic  speech  generation.
Primary  function :  Generate  spoken  audio  from  text  input.
Model  Versions:
ElevenLabs’  TTS  systems  may  be  deployed  in  multiple  model  versions  or  iterations  (e.g.,  v1,  v2,
v3).

These

versions

reflect

ongoing

development

and

may

differ

in

architecture,

scale,

and

training

data

composition

over

time.

Training  Data  Sources
Training  data  for  TTS  models  may  include,  across  different  model  versions,  a  combination  of:
●  Licensed  voice  recordings,  including  across  multiple  languages,  accents,  and  speaking
styles;
 ●  Publicly  available  data,  including  lawfully  accessible  speech  and  text  data;  ●  Proprietary  data,  including  datasets  created,  commissioned,  or  curated  by  ElevenLabs;
and

●  User-provided  data,  where  users  have  permitted  such  use  or  where  ElevenLabs
otherwise

has

a

lawful

basis

to

use

the

data

for

training

or

improvement

purposes.

Not  all  training  data  sources  are  used  in  every  model  version,  and  the  relative  contribution  of
each

source

may

vary

by

model

version.

Dataset  Scale  and  Characteristics
●  Datasets  may  include  large  numbers  of  voice  samples  across  multiple  languages,
accents,

and

speaking

styles.
 ●  Data  points  may  include  voice  recordings  and  associated  metadata  (e.g.,  language,
speaker,

or

recording

conditions).

Intellectual  Property  and  Personal  Data
Training  data  may  include  copyrighted  material  and  personal  information,  including  voice  data.
Voice

data

is

used

subject

to

applicable

privacy,

consent,

and

data

protection

requirements.

2.  Voice  Models  (Voice  Conversion  and  Voice  Cloning)
Operational  domain :  AI  audio  –  voice  transformation  and  voice  cloning.
Primary  function :  Transform  or  generate  speech  using  specific  voice  characteristics.
Training  Data  Sources
Training  data  for  Voice  models  may  include  a  combination  of:
●  Licensed  voice  recordings,  including  across  multiple  languages,  accents,  and  speaking
styles;
 ●  Proprietary  data,  including  datasets  created,  commissioned,  or  curated  by  ElevenLabs;
and
 ●  User-provided  data,  where  users  have  permitted  such  use  or  where  ElevenLabs
otherwise

has

a

lawful

basis

to

use

the

data

for

training

or

improvement

purposes.

Dataset  Scale  and  Characteristics
●  Datasets  may  include  voice  samples  across  multiple  languages,  accents,  and  speaking
styles.
 ●  Data  points  may  include  voice  recordings,  and  associated  metadata  (e.g.,  language,
speaker,

or

recording

conditions).

Intellectual  Property  and  Personal  Data
Training  data  may  include  copyrighted  material  and  personal  information,  including  voice  data.
Voice

data

is

used

subject

to

applicable

privacy,

consent,

and

data

protection

requirements.

3.  Speech-to-Text  (STT)  Models
Operational  domain :  AI  audio  –  transcription  of  speech  into  text.
Primary  function :  Convert  spoken  audio  into  written  text  or  structured  textual  representations.  Training  Data  Sources
Training  data  for  STT  models  may  include  a  combination  of:
●  Licensed  speech  recordings  paired  with  corresponding  text  transcriptions;  ●  Publicly  available  data,  including  lawfully  accessible  speech  and  transcription  datasets;  ●  Proprietary  data,  including  datasets  created,  commissioned,  or  curated  by  ElevenLabs;
and
 ●  User-provided  data,  where  users  have  permitted  such  use  or  where  ElevenLabs
otherwise

has

a

lawful

basis

to

use

the

data

for

training

or

improvement

purposes.

Dataset  Scale  and  Characteristics
●  Training  datasets  may  include  large  volumes  of  audio  data  and  corresponding  text,
expressed

as

ranges

or

estimates.
 ●  Data  points  may  include  audio  recordings,  text  transcriptions,  and  associated  metadata
(e.g.,

language,

speaker,

or

recording

conditions).

Intellectual  Property  and  Personal  Data
Training  datasets  may  include  copyrighted  content  and  personal  information,  including  voice
data.

Data

is

used

in

accordance

with

applicable

privacy

and

data

protection

laws.

4.  Music  Models
Operational  domain :  AI  audio  –  music  audio  generation.
Primary  function :  Generate  or  transform  music.
Training  Data  Sources
Training  data  for  Music  models  may  include  a  combination  of:
●  Licensed  music  recordings;   ●  Proprietary  datasets,  including  internally  created,  commissioned  audio,  or  internally
labeled

datasets

derived

from

or

based

on

licensed

music

recordings;

and
 ●  Synthetic  audio,  generated  to  supplement  training  or  evaluation.
Dataset  Scale  and  Characteristics
●  Dataset  sizes  may  be  expressed  as  ranges  or  estimates.  ●  Data  points  may  include  music  audio  recordings  and  associated  metadata.
Intellectual  Property
Training  data  for  these  models  may  include  content  protected  by  copyright  or  other  intellectual
property

rights

and

is

used

in

accordance

with

applicable

licensing

terms.

5.  Non-Speech  Audio  Models
Operational  domain :  AI  audio  –  non-speech  audio  generation.
Primary  function :  Generate  or  transform  other  non-speech  audio  content.
Training  Data  Sources
Training  data  for  Non-Speech  audio  models  may  include  a  combination  of:
●  Licensed  non-speech  audio  recordings,  including  sound  effects;   ●  Proprietary  datasets,  including  internally  created  or  commissioned  audio;  and  ●  Synthetic  audio,  generated  to  supplement  training  or  evaluation.
Dataset  Scale  and  Characteristics
●  Dataset  sizes  may  be  expressed  as  ranges  or  estimates.  ●  Data  points  may  include  audio  recordings  (e.g.,  sound  effects  or  ambient  audio)  and
associated

metadata.

Intellectual  Property
Training  data  for  these  models  may  include  content  protected  by  copyright  or  other  intellectual
property

rights

and

is

used

in

accordance

with

applicable

licensing

terms.

Substantial  Modifications
ElevenLabs  updates  its  training  datasets  and  models  on  an  ongoing  basis.  This  Disclosure  will
be

updated

upon

any

substantial

modification,

including

material

changes

to

training

data

sources,

dataset

composition,

or

model

performance

characteristics,

prior

to

or

concurrent

with

public

availability

in

California

and

other

applicable

jurisdictions.

Changes

in

training

data

composition

across

successive

versions

within

an

existing

model

family

do

not,

by

themselves,

constitute

a

new

model

family

unless

they

result

in

a

material

change

to

the

system’s

modality

or

primary

function.

 Limitations
This  Disclosure  is  intended  to  provide  meaningful  transparency  while  protecting  confidential
business

information,

trade

secrets,

and

security-sensitive

details.

Certain

information

is

therefore

provided

in

summarized

or

estimated

form,

as

permitted

by

law.

References

to

user-provided

data

in

this

Disclosure

are

subject

to

ElevenLabs’

applicable

privacy

notices

and

data

protection

obligations.

This

Disclosure

is

not

intended

to

create

contractual

commitments

or

expand

ElevenLabs’

obligations

beyond

those

required

by

applicable

law.

  Contact  Information
Questions  regarding  this  disclosure  may  be  directed  to:   Eleven  Labs  Inc.   Legal  &  Compliance  Team  Email:  legal@elevenlabs.io