Books

RedLectern topic

AI media and information

Synthetic media · misinformation · deepfakes · provenance · detection limits · media literacy.

For: Everyone · creators · journalists.

After this page you can: Assess media provenance without treating detection as proof.

Page reviewed

Part of: Artificial Intelligence › AI and society

About AI media and information

AI-generated and altered images, audio, and video can circulate with a misleading caption, an old context, or no explanation of how they were made. The useful question is not only “Is this synthetic?” Ask where the item first appeared, what the available record says happened to it, whether the account or publisher is credible, and whether independent evidence supports the claim being made. A practical media-literacy routine is to stop, investigate the source, find better coverage, and trace the media to its original context.

Provenance records and detection tools can contribute evidence, but neither settles whether a media claim is true. A signed record can help describe an asset’s history and whether that record is associated with the asset; it cannot establish that the depicted event happened as described. Detection results also depend on the tool and the material being examined, and errors are possible. For everyday viewing, start with source and context checks. For creators and journalists, document what was checked, distinguish confirmed facts from uncertainty, and explain what the evidence does and does not establish.

Start here

route

Start with four media-checking moves

Learn to stop, investigate the source, find better coverage, and trace media to its original context.

Start with: SIFT (The Four Moves)

Learn the foundations

1.What provenance can tell youProvenance is information about the history of a digital asset: for example, claims about its creation or later edits.

Provenance is information about the history of a digital asset: for example, claims about its creation or later edits. Content Credentials are a C2PA mechanism for associating signed provenance information with media. When a credential is available, check what it actually records, whether its signature and association with the asset validate, and who signed it. These checks help you understand the record and its source.

A valid provenance record is evidence about an asset’s recorded history, not a verdict about the truth of its contents. A genuine photograph can be paired with a false caption; a credential may be missing or incomplete. Read the record alongside the original source, context, and corroborating evidence. Treat watermarking as one possible way to carry a signal or recover a provenance record, not as proof that a claim is true.

Deepfake

Manipulated or synthetic audio or visual media that seems authentic and features a person who appears to say or do something they never said or did, produced using artificial intelligence techniques.

Content Credentials

A cryptographically bound record that records an asset’s provenance, such as its origin, modifications, or use of AI.

C2PA

An organization that develops technical standards for certifying the source and history of media content.

Watermarking

Embedding a detectable signal in content to indicate origin or ownership.

Try it

A photo has a valid signed credential saying it was cropped. What does that establish, and what would you still check before sharing its caption?

It supports that the credential is associated with the asset and records a crop action. It does not establish that the caption is accurate. I would check the original source, date and location, and independent reporting or other evidence for the event.

Deepfakes And Content Credentials

2.Check the source and contextWhen an image or clip makes a surprising claim, pause before forwarding it.

When an image or clip makes a surprising claim, pause before forwarding it. Trace it to the earliest source you can find, check the date and setting, and compare the claim with independent reporting or primary material. A reverse image search or a search using a video frame can help find earlier appearances or a different context. For journalism, note which parts of the chain are confirmed and which remain unclear.

The origin of a file and the truth of its caption are separate questions. A real recording can be old, edited, or presented as evidence of something else. A synthetic item can also be shared as satire or illustration rather than as a factual record. A useful routine is to stop, investigate the source, find better coverage, and trace the media to its original context. If the evidence is inconclusive, say so instead of turning uncertainty into a confident label.

Try it

A post says a video shows a public figure making an announcement today. List two checks you would make before repeating the claim.

I would search for the earliest upload or an earlier appearance of the clip, then check whether the date and event match the post. I would also look for an official recording or independent reporting that corroborates the announcement.

3.Treat detection as one clueA detector estimates whether the item matches patterns associated with synthetic or manipulated media.

A detector estimates whether the item matches patterns associated with synthetic or manipulated media. Its output is not a direct observation of how an asset was made. Detector performance varies with the method and evaluation material; compression, resizing, noise, and other changes can affect results. A result from one tool should therefore be described as a signal to investigate, not as a confirmation.

Visual clues can also be suggestive without being decisive. Missing artifacts do not establish that media is genuine, and an unusual detail does not establish that it is fake. Combine any detection result with provenance information, source checks, context, and independent evidence. If you report a detector result, identify the tool and the file tested, and explain the uncertainty. The aim is to assess what the available evidence supports, not to force every item into a real-or-fake verdict.

Try it

A detector labels a compressed video “likely synthetic.” Name one reason to treat that result cautiously and one independent check to make.

Compression can affect detector performance, so the result is not conclusive. I would locate the earliest available version and check the clip’s source and context against independent reporting or an original recording.

Latest

No recent updates are available.

See the sources list

Books

Glossary

4 terms on this page.

Browse the AI glossary

Questions and comments