FIELD GUIDE · DEEPFAKES, PROVENANCE & EVIDENCE

CAN YOU STILL PROVE ANYTHING ONLINE?

A photo, video, voice note, screenshot, or document can still be evidence. It just cannot carry the whole case by itself anymore.

UPDATED 2026-09-22

THE PROBLEM

PIC OR IT DIDN'T HAPPEN IS DEAD

The screenshot has entered its forensic era.

Synthetic media did not invent forgery, but it lowered the cost of producing convincing images, audio, video, and documents. Verification therefore shifts from asking only 'does this look fake?' toward asking where the file came from, what happened to it, who signed its history, and whether independent evidence agrees.

01

LOOK

SEEMS REALHuman intuition is now the weakest layer.
02

TRACE

WHERE FROM?Source and edit history matter more.
03

VERIFY

SIGNED?Credentials can show tamper-evident provenance.
04

CORROBORATE

WHAT ELSE?Reality still needs independent evidence.

THE FIGHT

WHAT COUNTS AS EVIDENCE WHEN ANYONE CAN GENERATE THE RECEIPT?

Courts, newsrooms, families, banks, employers, platforms, and ordinary people all use media as evidence. The danger is two-sided: fabricated media can falsely persuade people, while genuine evidence can be dismissed as 'probably AI' once synthetic media becomes common.

THE INFRASTRUCTURE CASE

PROVENANCE CAN RESTORE A LOT OF USEFUL TRUST

Open standards can bind tamper-evident creation and edit history to media, while watermarks and fingerprinting can help recover provenance after ordinary transformations. This gives publishers and consumers more context than visual inspection alone.

A passport does not prove every story you tell, but it gives identity claims a trusted chain.

THE SKEPTICAL CASE

NO SINGLE BADGE CAN PROVE THE WORLD BEHIND THE FILE

C2PA itself says provenance can show origin and history without proving that depicted content is factually true. Credentials may also be absent or removed, and detection systems face adversarial pressure.

A notarized photo of a staged scene is still a staged scene.

THE WEIRD SHIT

THE REAL FILE MAY NEED A PEDIGREE

THE PROVENANCE STANDARD EXPLICITLY DOES NOT CLAIM TO PROVE TRUTH

C2PA says Content Credentials verify provenance assertions and tamper evidence, not whether an image, video, or claim depicts factual reality.

Authenticity of the file and truth of the scene are different questions.

SYNTHETIC-CONTENT DEFENSE IS ALREADY MULTI-LAYERED

NIST's synthetic-content report surveys provenance, labeling, watermarking, detection, testing, and auditing rather than identifying one universal solution.

The technical answer is a stack of methods.

REAL-WORLD AUTHENTICITY SYSTEMS HAVE EDGE CASES

Microsoft Research's 2026 review of media integrity methods emphasizes practical limits, attack surfaces, and the need to combine provenance, watermarking, and fingerprinting.

Verification infrastructure must survive ordinary editing as well as adversaries.

MAJOR AI PROVIDERS ARE NOW SHIPPING PROVENANCE SIGNALS

OpenAI says supported generated media includes C2PA credentials and watermarking, with verification tools expanding across images and audio.

Creation-side provenance is moving from standards work into deployed products.

THE PEOPLE WITH A STAKE

WHO BENEFITS WHEN EVERYTHING CAN BE CALLED FAKE?

ORDINARY USERS

They face more work verifying urgent, emotional, or financially consequential media.

NEWSROOMS + COURTS

They need stronger chains of custody and corroboration while preserving the ability to accept genuine evidence that lacks modern credentials.

CAMERA + AI COMPANIES

They can embed provenance or watermarking at creation, but adoption and interoperability determine whether the signals survive real workflows.

BAD ACTORS

They benefit from both directions of confusion: fake media can look real, and real media can be dismissed as fake.

THE UNANSWERED QUESTION

WHAT BECOMES THE NEW DEFAULT PROOF?

TAKE THIS TO DINNER: Seeing is still evidence. It is no longer the final witness.

RECEIPTS

Provenance can authenticate a history. It cannot certify reality itself.

  1. C2PA and Content Credentials ExplainerCoalition for Content Provenance and Authenticity
  2. Reducing Risks Posed by Synthetic ContentNational Institute of Standards and Technology
  3. Media Authenticity Methods in Practice: Capabilities, Limitations, and DirectionsMicrosoft Research
  4. Advancing content provenance for a safer, more transparent AI ecosystemOpenAI