Intelligence
Readable explanations alongside deterministic detection results.
OUTPUT: Auditable, structured context
LEARN MORE INTRODUCING A NEW WAY TO ASSESS SYNTHETIC MEDIA
Deepfakes are everywhere. resemble helps detect them.
Examine audio, video and images with detection designed for explainable decisions.
Detect AI-generated audio, video and images with results teams can understand. Verify content, respond to fraud and make confident decisions without losing time.
Detection models usable anywhere deepfakes happen.
Contact center fraud
Integrate detection into a review workflow. Give your team context before a decision is made.
Get started Inbound audio call
Resemble analysis
◎ Identity
▥ Detect
◉ Signal
Illustrative alert
Call flagged for review
An analyst receives context and a recommended next step.
Suggested response
Pause the transaction and review the call.
EXAMPLE WORKFLOW · NOT A LIVE RESULT
Put detection where your team already works.
Explore ways to connect a detection workflow with calling, meetings and content review.
Explore integrations
Deploy detection with an API.
Start with a file, request an analysis and review a structured response.
detection-example.py
# Examine an audio file
from resemble import Resemble
with open("sample.wav", "rb") as audio:
response = Resemble.detect(audio)
print(response["assessment"])Built for teams working across media, communications and security
MEDIA TEAMS◈ CommunicationsENTERPRISE◎ Trust & SafetySECURITY✳ Developers
DETECTION MODELS
Generative AI is racing toward realism. Detection must keep pace.
Models for audio, image, video and text help teams assess the media they encounter, with findings they can put into context.
LATEST RELEASE
▥ DETECT-World
Assess AI-generated audio, video and images. DETECT-World combines pattern analysis with a world-model approach to examine departures from physical reality.
OUTPUT: Structured score, assessment and readable context
REFERENCE ACCURACY: Audio 99.5% | Video 98% | Image 96%
Three generations of detection
■ AUDIO■ IMAGE■ VIDEO
Audio shown across all three generations; image and video were introduced later. Reference figures vary by benchmark and test conditions.
Signal
Alert a reviewer when content resembles a known fraud pattern.
OUTPUT: A reviewable alert and notification
EXPLORE SIGNALPerTh Multimodal
Imperceptible watermarks across media formats.
OUTPUT: Provenance information for review
LEARN MOREIdentity
Enroll a voice or likeness and assess a subsequent identity match.
OUTPUT: A score, assessment and readable context
LEARN MOREBENCHMARKS
Third-party validated benchmarks
A benchmark is most useful when its methods are visible. Public leaderboards, sources and test conditions help put detection results in perspective.
Explore benchmarks99.5%
Audio detection accuracy
95.8%
Image detection accuracy
98.2%
Video detection accuracy
| # | Model | RTF | Accuracy |
|---|---|---|---|
| 1 | DETECT-World | 0.03 | 99.5 |
| 2 | DETECT-3B Omni | 0.08 | 98.2 |
| 3 | Whispeak | 0.39 | 97.70 |
| 4 | Aurigin AI | 0.33 | 96.75 |
| 5 | Pella Research | 0.021 | 95.82 |
| 6 | Pindrop | 0.076 | 95.05 |
| 7 | Corsound AI | 0.035 | 87.79 |
| 8 | Hive | 0.34 | 83.53 |
| 9 | Reality Defender | 1.52 | 71.27 |
| 10 | Wav2Vec2 (2019 LA) | 0.14 | 62.89 |
| 11 | AASIST | 0.017 | 56.83 |
| 12 | Wav2Vec2 (2024 mix) | 0.056 | 55.55 |
| 13 | Deepfake-V2 | 0.027 | 53.03 |
| 14 | AST | 0.003 | 50.99 |
| 15 | RawNet2 | 0.035 | 50.66 |
| 16 | LCNN-LFCC | 0.0056 | 50.00 |
| 17 | AASIST (2019 LA) | 0.11 | 48.17 |
| 18 | AASIST3 | 0.13 | 47.63 |
“
Responsible AI needs a bridge between powerful creation tools and the trust people need to use them.
Perspective from the responsible AI community
Responsible AI
“
Detection has to move quickly as synthetic media changes. Clear results make that work easier to act on.
Perspective from detection teams
Detection teams
“
As generated voices become more convincing, verification becomes part of everyday communication.
Perspective from communications teams
Communications teams
SECURITY AND COMPLIANCE
Enterprise-grade, multimodal, available on-prem.
For environments where security is non-negotiable, consider deployment, retention and regulatory requirements together.
01GDPR
Plan data handling around a lawful basis and appropriate protections.
02ISO 27001
Evaluate information security management requirements.
03Deployment
Choose a rollout path that fits your infrastructure and review process.
04API-first
Connect detection to existing security and identity workflows.
05SOC 2
Ask how availability and confidentiality controls are assessed.
06HIPAA
Review healthcare data-handling needs before deployment.
07Air-gapped
Explore on-premises options for isolated environments.
08Retention
Set media-retention rules to match your requirements.
REPORTED INCIDENTS
Deepfake incident database
A sample of reported incidents. Many more go unreported. See the kinds of cases that have become public.
Explore reported incidentsAI-generated intimate material reported after a relationship ended
A public account describes the personal harm that can follow synthetic media made without consent.
Restaurant worker reports a voice impersonation attempt
An impersonator reportedly contacted a manager and appeared to speak for an employee.
Online post raises questions about a manipulated video
Viewers encountered footage whose authenticity required closer examination.
Synthetic voice used in a family emergency scam
A familiar-sounding message reportedly prompted an urgent request for money.
Reported messaging scam involved AI voice notes
Voice messages presented as a relative were reportedly used to request transfers.
Deepfake advertisement prompts an investor fraud warning
A manipulated promotional video was reportedly connected to an attempted transfer.
Social video impersonates a public figure
An account reportedly circulated synthetic clips to solicit money from followers.
Fraudulent audio shared during an urgent request
A convincing voice alone was not enough to establish who sent the message.
RESOURCES
New to this? Here’s where to start.
The 2026 Deepfake Threat Report
Explore patterns in synthetic media attacks, reported fraud and the questions facing enterprise defenders.
VIEW REPORT
Introducing a world model for deepfake detection
Learn how checking the behavior of media can complement the search for familiar generation artifacts.
VIEW RESEARCH
A practical guide to synthetic media disclosure
Understand the questions providers and deployers should ask about labeling, provenance and disclosure.
VIEW GUIDEKnow what’s real — and what’s a real threat.
Explore detection for the audio, video and images your team encounters, and build a clearer path from finding to decision.
Try detection