PNG’s answer to the deepfake problem

TheNew York Times (Opens in new window)recently reported the world’s leading deepfake expert Hany Farid no longer trusted his own eyes. Farid, a University of California professor in Berkeley, felt he was “going blind”. Evolving at breakneck speed, AI had obscured the truth, distorted reality and fractured democracies. The Times noted Farid’s earlier research findings that many people “could no longer distinguish a real photograph from a digital creation, a real voice from an AI clone, a real video clip from a wholesale fabrication”.

It is a long stretch from the advanced technology economy of Berkeley to PNG where resources are comparatively meagre and high-end cyber expertise scarce. Financially constrained media in PNG generally operate on a small scale with limited editorial and production capacity – they cannot afford to dedicate ample staff time for forensic analysis or deep dives into social media burrows. Yet the communities they serve are highly vulnerable to manipulation and falsehoods amplified through digital media.

Almost 40% (Opens in new window) of the PNG population of an estimated 11 million lives in poverty, lacks access to basic services or employment in the formal economy, is poorly educated and possesses very low media literacy. Millions of these same people have access to social media applications such as Facebook via mobile telephones. Many are beyond the reach of mainstream journalism and its espoused principles of accuracy and accountability.

PNG’s vulnerability to the scourge of mis- and disinformation is mirrored elsewhere in the Pacific as noted by the most recent State of the Media (Opens in new window)report. This regional challenge of the digital “coconut wireless (Opens in new window)” was covered in The Interpreter by Michaela Long and Connor Graham. PNG is a test case. Of specific concern, it prepares for national elections in 2027 fearing a recurrence of the violence (Opens in new window) last time, which resulted in more than 400 deaths and the displacement of some 20,000 residents.

In March 2026, a parliamentary committee (Opens in new window) report in PNG responded emphatically to the concerns of the political class and mainstream media. The Inquiry into the Standard and Integrity of Journalism in PNG made 40 recommendations, 20 of which involved – and served to endorse – the self-regulating role of the Media Council of PNG (Opens in new window). Among those recommendations was that public funding be provided to the volunteer-run Media Council in support of its independent processes of industry self-regulation and development.

The report proposed a national disinformation taskforce involving the Media Council and other cross-sectoral bodies, and called on the Council to create industry-wide fact-checking protocols and verification standards. It argued government should lead development of a national digital and media literacy strategy, introduce a Freedom of Press Act, and amend the PNG Criminal Code to help protect at-risk media workers. If implemented, the recommendations would help create a substantial scaffolding for civic and social stability.

But they would still be insufficient to help media uphold “human-in-the-loop” editorial standards, or to serve as a model for other national jurisdictions struggling similarly. It is not just that AI creations and deepfakes are becoming so difficult for human editors to detect or that many activities of fact-checking and verification occur when suspicious content has already taken hold in the community. So long as organisations lack adequate capacity to perform systematic and time-consuming verification work, training and the setting of verification protocols will offer only part-remedies. Nor will mainstream media narrow the information gap with the millions of social media users who do not consume their services.

The Media Council of PNG has proposed a co-designed technology response that, regrettably, has received scant attention from policymakers.

With an eye to the 2027 PNG elections, the Council together with computer science researchers from the University of New South Wales (UNSW) and the University of Technology Sydney (UTS) proposed an editorial aid using advanced technology applications already developed and tested successfully. Thisautomated fake news detector, trained specifically for the PNG civic and media context, would provide near-real-time alerts of suspicious multimodal content, including text, images, video and synthetic audio. It would provide bilingual (English, Tok Pisin) summaries and explanations, and also assign different risk levels to distinguish normal political exaggeration from factually incorrect and high-risk deepfakes or harmful manipulation. An optional plug-in would allow authorised fact-checkers to issue advisory alerts to social media users.

The tool would combine a lightweight machine-learning detector for rapid screening with a complementary Large Language Model (LLM) for contextual reasoning and explanation. Designed for a specialised rather than universal purpose, it would minimise the factual errors and inconsistencies that still affect general-purpose LLMs.

This is no substitute for the human tasks of investigative and attributable reporting. At a minimum, the automated fake news detector would offer early alerts and advice as the basis for editorial decisions to publish or not, to identify trending mis- or disinformation, and provide rapid response opportunities for relevant authorities to attempt corrective actions. Whether for deployment in PNG or elsewhere, it is an inventive model deserving of serious evaluation in the interests of fact-based public discourse and democratic order.