Nobel Forensic Technologies / AI Investigations

AI-Generated &
Manipulated Media
Investigations

Forensic examination of images, video, audio, and documents suspected of AI generation or manipulation — using artifact-based methodology developed through a formal AI forensics program at DHS/FLETC.

Scope and limitations are defined explicitly for every engagement. No engagement promises a universal authenticity certification.

Discuss Your Matter
Methodology

How These Examinations
Are Conducted

AI forensics examinations rely on observable, documentable forensic artifacts — not automated AI detection tools. Commercial "AI detectors" produce results that are unreliable, untestable, and inadmissible as the sole basis for expert opinion.

Examinations draw on metadata analysis, compression signature inspection, file-level artifact review, pixel-level statistical analysis, and GAN or diffusion model artifact patterns — grounded in the same forensic methodology applied to all digital evidence.

This approach was developed through the AI forensics program built at DHS/FLETC addressing investigations involving AI-generated images, video, and audio across federal law enforcement contexts.

Images
Metadata, compression artifacts, pixel-level analysis, GAN and diffusion model signatures
Video
Frame-level artifact analysis, encoding inconsistencies, deepfake model signatures
Audio
Spectral analysis, voice cloning artifact patterns, encoding metadata
Documents
Metadata analysis, formatting artifacts, creation history
Background
AI forensics program developed and taught at DHS/FLETC

On automated AI detection tools: Tools marketed as "AI detectors" produce probabilistic outputs that cannot be independently validated and are not tied to observable forensic artifacts. This practice does not present their output as a standalone basis for any conclusion. Where such tools are used as one reference among many, that use and its limitations are disclosed in the report.

Scope & Limitations

What Every Engagement
Makes Explicit

The scope and limitations of each engagement are defined before examination begins and stated in the final report.

No Universal Certification
No examination can certify that media is definitively authentic or definitively AI-generated. Conclusions are expressed in terms of findings and their evidentiary weight — not absolute ground truth.
Tool Evolution
AI generation tools evolve rapidly. Artifacts detectable today may be absent in future model versions. The report states the state of knowledge at the time of examination.
Inconclusive Findings
Some examinations will not yield a definitive finding. High-quality AI-generated media or heavily processed authentic media may produce inconclusive results. Inconclusive findings are reported honestly.
Original File Required
Examinations require the original digital file where possible. Screenshots or re-encoded copies degrade forensic artifact quality and limit conclusions. This limitation is stated in the report when it applies.
Subject to Challenge
As with all expert opinion, findings are subject to cross-examination. The methodology used here is documented and reproducible — because that is the standard that holds up.
Applications

Where This Work
Arises in Practice

Litigation & Civil Matters

Authentication disputes over images, video, or documents submitted as evidence. Examination of media alleged to have been fabricated or altered using AI tools.

Criminal Defense & Prosecution

Digital media introduced as criminal evidence — images, surveillance footage, and communications — examined for signs of AI generation or manipulation.

Defamation & Harassment

Examination of images or video alleged to be AI-generated deepfakes used in defamation, harassment, or non-consensual intimate image matters.

Insurance & Fraud

Photographic or video evidence in insurance claims and fraud investigations where AI manipulation of damage, injury, or scene documentation is alleged.

$300
Per hour · Written scope · Budget approval required

Scope and budget are defined in writing before examination begins. Related: Investigating potentially AI-generated images, video, and audio →

Discuss an AI forensics
engagement.

Discuss a Matter