Generative AI can add plausible details and confident narratives to incomplete recordings, making weak evidence appear stronger without new measurements, as illustrated by hypothetical “ectoplasm” restorations and real concerns in medical imaging.

International efforts—NIST’s forensic deepfake testing, ONVIF’s candidate signed-video standard, JPEG Trust provenance work and EU AI transparency rules—can verify integrity and processing history but cannot certify what a camera actually captured or that a model-generated feature is real.

Investigators must preserve original files, label every enhancement, distinguish ordinary corrections from generative additions and test claims with independent physical or observational evidence rather than treating AI output as a second observation.

From historical ectoplasm photographs to AI-enhanced video, convincing details can come from a model rather than a camera. New international authenticity efforts expose the difference.

Imagine a dark, blurry recording in which a pale shape seems to move beside a doorway. A viewer asks an AI tool to sharpen it. The result shows what looks like a human face. A chatbot then supplies a confident explanation of what the face might be. No one has taken a second photograph or found a new witness. Yet the story suddenly looks better documented than it did a minute earlier.

That scene is hypothetical. Its underlying problem is not. Generative restoration can create plausible detail from incomplete information, while fluent machine descriptions can make interpretation sound like observation. The danger is not that artificial intelligence has found a ghost. It is that AI can make a weak piece of evidence appear stronger without adding a new measurement.

This is becoming a live international standards issue. The U.S. National Institute of Standards and Technology updated its Guardians of Forensic Evidence initiative on October 30, emphasizing the need to test deepfake tools against the compressed, edited and imperfect images investigators actually receive. In August, the global camera-standards organization ONVIF released a candidate standard for cryptographically signed video. In Europe, AI-generated-content transparency rules began applying in August. Their shared challenge is not whether pictures can look convincing. It is whether investigators can establish what a camera captured and what later processing supplied.

The old photograph and the new model

The Victorian and twentieth-century fascination with ectoplasm offers an unusually clear........

© Eurasia Review