Oman put AI inside a media excellence program. Test credibility under deadline

THE SIGNAL IN ONE SENTENCE

Credibility is easy to promise when nobody is waiting for the bulletin. The harder test starts when a video lands twelve minutes before air, an editor wants three platform versions, the source is slow to answer and an AI tool offers a polished shortcut. On October 25, Oman's Ministry of Information held a lecture at its theatre titled Future Media: Creativity and Innovation in the Era of Artificial Intelligence. Oman News Agency reported that Abdulwahid Al Ghanem, an institutional innovation specialist and consultant, delivered the session. The stated aim was to help participants understand the relationship between creativity, innovation and AI in media. The lecture encouraged them to turn media opportunities into innovative, testable ideas and to use AI responsibly while preserving authenticity and credibility. The event sat inside the accompanying activities of the Ministry's Media Excellence Competition. The published themes included AI as an enabler of creativity, future media technologies, authenticity, trust, responsibility and future media skills. That combination is useful because it refuses the lazy choice between innovation and credibility. A newsroom should be able to use new tools without handing its standards to them. But a lecture about responsibility is not evidence that a workflow remains responsible under pressure. The public account does not describe a newsroom simulation, publish a curriculum, name the AI tools participants used, report how many people attended or measure whether anyone became better at verifying generated material. It is an official description of one lecture within a competition program, not a national media policy or an evaluated intervention. The plain signal is this: test credibility under deadline. Give a team a realistic assignment, a short clock and a bundle of material that contains both useful evidence and traps. Ask them to produce a publishable item. Then inspect the chain between the first file and the final audience-facing claim. The exercise should not be a trivia quiz about spotting an obviously melted hand in an AI image. Modern newsroom risk is more ordinary than that. A generated transcript can omit a qualifying sentence. A translation can make a cautious statement sound certain. A summary can merge two speakers. A synthetic voice clip can imitate the cadence of a real official. A caption can place a genuine image in the wrong city. A headline assistant can turn a preliminary estimate into a confirmed total because the stronger wording wins attention. The asset may be partly real, partly transformed and completely misleading. Start the drill with a clear assignment. A weather disruption, public-service announcement, business result, cultural event or local emergency will do. Supply a press release, two interview clips, a photograph, a social post, an old background story and one AI-generated item. Include a contradiction that matters. Do not tell participants which item is synthetic. The team has fifteen minutes to decide what it knows, what it does not know and what it can responsibly publish. AI tools may transcribe, translate, search within documents, propose questions, summarize material or create platform variants. Every use has to leave a receipt. That receipt needs five things. First, preserve the original inputs. Save the source file, URL, capture time and person who obtained it. A screenshot without a link or timestamp is a dead end. A downloaded video without its posting context can be genuine footage attached to a false claim. If a file arrived through a messaging app, record who supplied it and what they say it depicts. Second, record what the AI changed. Did it remove silence, translate Arabic to English, clean audio, extend an image, generate a voice, summarize a document or draft the entire item? The label should describe the operation instead of hiding everything behind the word assisted. Third, tie every publishable claim to evidence. A source ledger can remain simple: claim, supporting item, verification action, uncertainty and editor. If a sentence says an event happened at a particular place and time, the row should show which source establishes both. If the evidence only suggests the location, the copy should not pretend otherwise. Fourth, keep a named human signoff. The final editor must have enough time, access and authority to reject the item. Clicking approve after reading only the generated summary is not oversight. The reviewer should be able to open the source, see the machine transformation, compare the wording and stop publication. Fifth, prepare the correction before the mistake. Who can pull the item? Which versions need correction across television, radio, website and social channels? How will the audience see that a change was made? Who contacts the affected source? A fast newsroom needs a fast repair path as much as a fast publishing path. The drill becomes useful when the clock tightens. At minute five, add a new clip that appears to confirm the story but has no known origin. At minute eight, make the named source unavailable. At minute ten, reveal that the translated quotation contains an ambiguous word. At minute twelve, ask for a vertical video and a push alert. Watch what the team sacrifices first. Does the source note disappear? Does uncertainty vanish from the headline? Does the editor approve a translation they cannot assess? Does the social version acquire a claim that the main story never verified? Does the group disclose the synthetic asset but forget that a genuine photograph was cropped in a way that changed its meaning? That is the point of pressure testing. Standards often fail at the handoff, not in the policy document. Score the exercise in layers. Accuracy comes first. Did each factual statement match the available evidence? Did the team distinguish direct observation, source testimony, official claims and its own inference? Provenance comes next. Could another editor reconstruct where each asset came from and what happened to it? The Coalition for Content Provenance and Authenticity has developed Content Credentials as a technical way to record an asset's origin and editing history with tamper-evident information. That can help preserve context across tools. It is not a truth machine. C2PA's own explainer says provenance can help establish origin, history and authenticity, but cannot by itself determine whether content is true or factual. A perfectly signed false statement is still false. A missing credential does not automatically prove deception. Technical provenance should strengthen reporting, not replace it. Then score disclosure. Could the audience understand when AI materially contributed to an image, voice, translation or text? A disclosure should be specific enough to be useful and short enough to survive publication. Generated illustration is more informative than enhanced. AI-translated interview, checked by a named editor, says more than technology assisted. Then score restraint. Did the team hold an item when the evidence remained weak? Did it publish a narrower claim instead of filling the gap with probability and vibes? A media excellence program should reward the person who removes an unsupported sentence, not only the person who ships the flashiest package. Finally, score repair. Give the team proof that one detail is wrong after publication. Measure how long it takes to correct every version, notify the audience and update the internal record. A correction hidden in the source file while the wrong clip keeps circulating is not a completed correction. This approach fits the language of the Omani lecture without pretending to be its curriculum. Testable ideas should be tested against failure. Responsible use should identify an accountable person. Authenticity should describe origin and transformation. Credibility should be earned through verification and visible correction. Future skills should include the ability to stop a sophisticated tool when it is producing the wrong thing quickly. UNESCO's global ethics recommendation offers a compatible baseline. It emphasizes transparency, traceability, human oversight and ultimate human responsibility. Its journalism work likewise stresses that AI can support professional work but cannot replace editorial responsibility or the public interest. The local version still matters. An Omani drill should reflect the languages, institutions, platforms and source relationships of Omani media. It should test Arabic transcription and translation, names and titles, regional accents, official terminology, archive footage, mobile-first distribution and the pressure to publish simultaneously across several formats. It should include independent, institutional, community and official sources rather than assuming credibility comes from one channel alone. The exercise can respect newsroom roles while still asking who has permission to challenge a claim and stop publication. Competition entries offer a natural laboratory. Participants could submit not only the final story, video or broadcast segment, but also a compact evidence packet: original sources, AI-use log, transformation history, verification ledger, signoff and correction plan. Judges could evaluate the finished work alongside the process that made it trustworthy. That would change the incentive. Without a process score, the most polished entry can hide the most fragile evidence. With one, creativity includes designing a workflow that remains legible under pressure. The results could be published in aggregate. Which failure appeared most often? Translation drift? Missing source context? Weak disclosure? Unsupported headlines? Unclear ownership? Slow correction? Which controls prevented an error without making the workflow unusably slow? Do not publish private source material or turn participants into examples of shame. Publish the pattern, the fix and a reusable drill. The result could be small enough to use next week: one scenario pack, one source ledger, one transformation log, one signoff card, one correction checklist and one scoring rubric. Run it once with experienced editors. Run it again with students and creators. Change the platform, format and language. Keep the underlying standard steady. The Ministry lecture put the right words in the same room: creativity, AI, authenticity, trust and responsibility. Now put them on the same clock. A newsroom does not prove credibility by saying it values credibility. It proves credibility when the timer is red, the clip is convincing, the source is unavailable and someone still asks for the receipt before pressing publish.