witness
LEGAL

Is this evidence real?

Screen digital evidence for AI generation before it enters proceedings. A strong first signal, not a replacement for expert testimony.

THE PROBLEM

Digital evidence is increasingly unreliable.

AI can generate realistic photos, videos, and audio that are indistinguishable to the human eye. Courts are starting to address this with proposed rules like FRE 707 and amendments to FRE 901(c). Attorneys need screening tools to evaluate evidence before relying on it or presenting it.

96%
Of people cannot reliably identify AI-generated faces
400%
Increase in deepfake incidents since 2023
FRE 901(c)
Proposed amendment for digital evidence authentication
32 states
Have introduced deepfake-related legislation
PRACTICE AREAS

Where it applies.

Civil Litigation

Screen photographic and video exhibits before filing. Identify potentially AI-generated evidence submitted by opposing parties. Flag manipulated media early in discovery to avoid costly surprises at trial.

Family Law

Evaluate the authenticity of digital evidence in custody disputes, protective orders, and divorce proceedings. Screenshots, photos, and videos submitted as evidence can now be fabricated with consumer-grade AI tools.

Criminal Defense

Challenge prosecution evidence that may be AI-generated or manipulated. Provide defense teams with a preliminary screening tool to identify exhibits that warrant further forensic examination.

Corporate Investigations

Screen whistleblower evidence, internal complaint documentation, and digital communications for signs of AI generation. Triage large volumes of digital evidence during internal investigations.

WHAT WITNESS CHECKS

Five layers of analysis.

01

AI-generated images

Ensemble detection models analyze pixel-level patterns, frequency artifacts, and generative fingerprints consistent with diffusion models, GANs, and other image generators.

02

Face swaps in video

Frame-by-frame analysis detects face replacement artifacts, temporal inconsistencies, and blending boundaries that indicate a face has been swapped onto another person.

03

Manipulated video

Detection of splicing, frame insertion, and temporal manipulation. Analysis of compression artifacts and inter-frame consistency.

04

C2PA provenance metadata

Verification of Content Credentials (C2PA) signed provenance data when present. Checks whether media carries a cryptographic chain of custody from capture device to current file.

05

EXIF and metadata analysis

Examination of file metadata for inconsistencies: mismatched timestamps, missing camera data, software signatures consistent with AI generation tools, and signs of metadata stripping.

IMPORTANT LIMITATIONS

What Witness is and is not.

Witness is a screening tool. It identifies statistical patterns consistent with AI generation and manipulation. It returns confidence scores, not certainties.

For proceedings requiring expert testimony, pair Witness results with a qualified digital forensics expert. Witness can help you decide which evidence to send to that expert.

We do not currently produce Daubert-standard reports. We do not claim court admissibility. We do not replace chain-of-custody procedures or forensic examination protocols.

Witness is a strong first signal. It tells you where to look harder.

HOW IT WORKS

Upload. Analyze. Decide.

01

Upload evidence

Submit images or video through the Witness web tool. Drag and drop or paste a URL. No software installation required.

02

Multi-model analysis

Multiple detection models run in parallel. Each checks for different indicators: generative fingerprints, frequency artifacts, face swap boundaries, metadata inconsistencies.

03

Review results

Receive a confidence score (0-100) and per-model breakdown. See which specific signals were detected and at what confidence level.

04

Inform your investigation

Use results to decide next steps. Flag evidence for forensic expert review. Document screening results as part of your due diligence process.

FAQ

Common questions.

Can AI-generated evidence be detected?

AI-generated images and videos leave statistical patterns that detection tools can identify. Witness screens media for these patterns using multiple detection models, returning a confidence score. However, detection is probabilistic, not deterministic. Results should inform further investigation, not replace expert testimony.

Is Witness admissible as evidence in court?

Witness is a screening tool, not a certified forensic instrument. It does not currently produce Daubert-standard reports. Results are best used to identify which evidence warrants further scrutiny by a qualified digital forensics expert who can provide testimony.

What types of digital evidence can Witness analyze?

Witness analyzes images for AI generation, videos for face swaps and manipulation, and checks for C2PA provenance metadata and EXIF data inconsistencies. It supports common formats including JPEG, PNG, MP4, and WebM.

How should I use Witness results in my practice?

Use Witness as a screening step during evidence review and discovery. If Witness flags media as potentially AI-generated, engage a qualified digital forensics expert for further examination. Document the screening as part of your due diligence. Do not present Witness results as forensic conclusions.

Can opposing counsel challenge Witness results?

Yes. Witness provides a preliminary screening signal, not a forensic conclusion. This is why we recommend pairing Witness screening with expert forensic analysis for any evidence that will be presented in proceedings. The screening helps you decide where to invest in that deeper analysis.

Screen your next case.

Talk to us about using Witness for digital evidence screening in your practice.