According to the study, this disconnect has created a growing divide between technical promises and practical news gathering. Many detection products and deepfake checkers are built in isolation. They often lack direct input from the journalists or platform engineers who would actually use them. As a result, expensive products often fail to help with daily editorial work.
Why confidence scores and black boxes fail working reporters
A major issue with these automated systems is how they present evidence to reporters under deadline pressure. The Princeton report notes that confidence ratings from AI image and video detectors are not useful to most journalists. Readers and editors typically view authentication as a binary question: an image is either real or it is fake. A probabilistic score does not give a reporter the certainty required to publish a correction or clear a story.
These detectors also function as unexplainable black boxes. When a tool flags a piece of media, it rarely explains why the algorithm reached its conclusion. Without clear provenance or inspectable evidence, reporters cannot defend their editorial decisions to the public or verify the findings independently. The report points out that until software creators work directly with reporters, automated verification tools will remain on the sidelines of major breaking news operations.
A confidence rating sounds precise at first. In practice, it is a liability for a reporter who needs a simple yes or no before publishing. If an editor cannot explain to a reader where a score came from, the rating carries no weight in the newsroom. The lack of direct journalist participation during the software design phase means developers keep building features that look good in lab demos but fall apart under deadline pressure.
Software hurdles and safety risks for sources on the ground
Technical friction inside newsroom publishing systems also limits these new authentication standards. Launched in 2021 by Adobe, Microsoft, and the BBC, the C2PA standard uses cryptographic manifests to record the history of media files. The goal is to embed these origins directly into the content. C2PA has gained support from major news organizations, including the Associated Press and The New York Times. Yet making the system work across standard newsroom software remains difficult.
According to one engineer from a North American newsroom, basic photo ingestion and editing software stripped C2PA manifests from incoming files. According to the report, the organization had to build custom developer patches just to keep the credentials intact. Worse, content management systems (CMS) - the software used to write and publish stories - often delete these credentials automatically upon publication.
Social media platforms add another layer of trouble. Many platforms strip out C2PA metadata and cryptographic signatures from images and videos when distributing stories. That erases the provenance trail before media reaches readers. Human rights group WITNESS says these digital signatures can be dangerous. In hostile regions, they may reveal a source’s location or camera serial number to the wrong people. Some frontline photographers have chosen cameras that lack the standard entirely.
Grounding verification tools in actual newsroom practice
Overcoming these barriers requires software developers to build tools around real editorial routines rather than theoretical models. When technology providers build alongside working journalists, they can avoid the black-box traps and metadata failures that derail current industry initiatives. Keeping tools transparent and grounded in daily reporting protects both the newsroom and the people it covers.
For newsrooms of all sizes, the path forward depends on practical verification systems that support human reporting without introducing new safety hazards or technical dead ends. As AI-generated content spreads online, news organizations need open, transparent tools that fit existing publishing pipelines and protect vulnerable sources on the ground.
Written by Dominik Czarnota using the Tribune Desk AI platform. Every claim in this article was fact-checked against its sources, and an editor read, edited and approved it before publication.
