Asia Centre finds Asian newsrooms are using AI with caution

Public-interest newsrooms across Asia are taking a cautious, piecemeal approach to artificial intelligence, according to a new Asia Centre report produced with DW Akademie and supported by the German Federal Ministry for Economic Cooperation and Development. The study covers independent and public-interest media in Afghanistan, Bangladesh, Bhutan, Cambodia, Indonesia, Myanmar, Pakistan and Uzbekistan. Rather than replacing journalists, AI has mostly been folded into routine support work. The report says outlets use it for translation, transcription, summarisation, early-stage ideation and final language editing. The Diplomat, which highlighted the findings, said generative tools were also being used for document summarisation, social media adaptation and headline polishing, while original field reporting, investigative drafting and editorial judgment remained human-driven. That limited use reflects what the report calls “principled caution”. More than 60% of surveyed newsrooms said adoption still happened through ad hoc experimentation by individual staff rather than a coherent editorial strategy. According to DW Akademie’s June announcement, respondents linked that caution to fears of factual errors, bias and damage to credibility. The report also points to weak internal systems as a major brake on more deliberate adoption. It highlights gaps in governance, training and strategic leadership, saying that individual AI literacy often fails to become organisation-wide policy or enforcement. Funding instability and resource constraints, it adds, make long-term investment in ethical and responsible AI difficult. Structural risks go beyond management. The report says the absence of formal editorial guidelines, missing data protection protocols and reliance on Big Tech platforms leave independent media exposed to algorithmic bias, privacy violations, copyright risks and state surveillance. It also says restrictive political and legal environments, together with regulatory ambiguity, impede responsible AI adoption across the region. Those pressures are compounded by severe language constraints in low-resource Asian languages. The Diplomat reported that major large language models underperform in languages including Bahasa Indonesia, Bangla, Burmese, Chittagonian, Dzongkha, Farsi, Jingpo, Khmer, Pashto and Urdu. The models often mistranslate context-sensitive terms, flatten cultural nuance and produce wrong outputs that still require labor-intensive human correction. Translation is the most widely adopted AI task, but outputs require labor-intensive human correction. The report also singles out exiled media, saying digital surveillance, internet blackouts, cross-border security threats and severe financial instability leave those outlets with little bandwidth to experiment. It argues that securely tailored tools for transcription, translation and media monitoring could help them reach diaspora audiences and report on repressed homelands without compromising journalists’ safety. Dr James Gomez, the Asia Centre’s regional director, said public-interest journalists wanted AI “only on terms that allow them to do their work with integrity.” Hélène Champagne, a DW Akademie program director and an editor on the report, said the path towards AI integration requires “a systemic approach” that accounts for journalists’ real constraints. The report concludes that long-term media sustainability will depend on local AI infrastructure, enforceable newsroom governance and open-source language models designed for the region’s political and linguistic realities.