On 7 April 2026, while LEAP’s halls sat empty 20 kilometres away in Malham and Iranian drones tested the Kingdom’s air defence systems overhead, 200 academics, journalists, and media professionals gathered at King Saud University in Riyadh for the 10th International Conference on AI in Media. The event — organised by the Saudi Association for Media and Communication, sponsored by KSU’s acting president Prof. Ali Masmali — proceeded without postponement, without relocation, and without the international audience that the Kingdom’s larger technology events demand. It was, in that sense, the most honest AI event Saudi Arabia hosted in 2026: domestic, professional, and focused on questions that the bigger conferences — with their $14.9 billion investment announcements and their celebrity CEO keynotes — rarely address.
The conference’s title — “Artificial Intelligence Media: Opportunities and Challenges” — is generic. Its substance was not. Across six specialised sessions, the participants examined how AI is restructuring the production, distribution, and consumption of media content in Arabic — a language spoken by 350 million people, written right-to-left, divided into dozens of regional dialects, and served by an AI ecosystem that is, by every measure, approximately three years behind the English-language AI frontier.
The timing was deliberate. Saudi Arabia designated 2026 as the Year of Artificial Intelligence. HUMAIN had launched ALLaM 34B — a 34 billion-parameter Arabic-first language model trained on approximately 3 trillion mixed Arabic and English tokens, per the model’s published research paper. HUMAIN Chat, the consumer-facing application built on ALLaM, was available on web, iOS, and Android. The Saudi Media Award had introduced the world’s first category dedicated to AI-generated content at the Saudi Media Forum in February — submissions opened 21 November 2025, closed 1 January 2026, and awards were presented on 4 February 2026 with criteria requiring explicit AI disclosure, artistic vision, and quality of execution. The infrastructure for AI-mediated Arabic media was being built. The conference asked what it would be used for.
The Arabic AI Gap
The Arabic AI market is where the English AI market was approximately three years ago — a characterisation used by multiple researchers at the conference and in the broader Arabic NLP community. The tools are catching up. The demand is enormous. The competition is thin. And the challenges are structurally different from those facing English-language AI.
Arabic’s computational complexity exceeds English by several dimensions. The language’s morphological richness — where a single root can generate dozens of derived forms through patterns of prefixes, suffixes, and internal vowel changes — makes tokenisation, the foundational step in language model training, significantly harder. The diacritical marks that disambiguate meaning in formal Arabic are frequently omitted in informal text, creating ambiguity that models must resolve through context rather than explicit notation. The right-to-left writing system, combined with the embedding of left-to-right numerals and Latin-script terms, creates rendering and processing challenges that monolingual English models do not encounter.
The dialect problem compounds the complexity. Modern Standard Arabic — the formal register used in news broadcasts, government communications, and academic writing — is no one’s native dialect. It is a learned register, analogous to Latin in medieval Europe. Actual speech and informal writing use regional dialects — Egyptian, Gulf, Levantine, Maghrebi — that differ from each other in vocabulary, grammar, and pronunciation as much as Spanish differs from Portuguese. A language model that understands MSA may fail entirely on Egyptian Arabic text, and vice versa. ALLaM addresses this by training on Saudi, Egyptian, Jordanian, and Lebanese dialects alongside Classical Arabic — but the challenge of comprehensive dialect coverage across 22 Arabic-speaking countries remains substantially unsolved.
The data scarcity problem is being addressed but not resolved. ALLaM’s approximately 3 trillion training tokens — mixed Arabic and English — is large by Arabic standards but modest compared to the frontier English models now training on 15 trillion tokens or more. The quality of available Arabic text data is uneven: news articles and government documents are well-represented, but informal text (social media, forums, messaging) and specialised domains (medical, legal, technical) are under-represented. HUMAIN has emphasised that the ALLaM team includes a meaningful proportion of Saudi PhD researchers and a deliberate gender mix — an investment in human capital that most Arabic-speaking countries cannot replicate at the same scale.
What the Sessions Revealed
The conference’s six sessions mapped the intersection of AI and Arabic media across production, distribution, ethics, and education. The discussions — reported by Arab News and Saudi Shopper — revealed both the enthusiasm and the anxiety that AI generates in a media ecosystem where the state controls the dominant platforms.
SAMC vice president Jareh Al-Marshidi, chairman of the conference’s scientific committee, described AI as “widely used across media workflows, from scriptwriting to distribution and audience targeting.” His emphasis was on maintaining “professional and ethical standards” — language that in a Saudi context carries a specific meaning: the ethical standards are not merely professional norms but regulatory requirements in a media environment where the state defines the boundaries of acceptable content.
Prof. Mutlaq Al-Mutairi, supervisor of the Dr. Ibrahim Al-Muhanna Chair for Energy and Specialised Media, identified limited research capacity and bureaucratic systems as obstacles to media AI innovation — a rare public acknowledgement that institutional constraints, not just technology gaps, slow Saudi AI adoption in media.
Ahmed Al-Dayhani, correspondent for Monte Carlo Radio in Saudi Arabia, argued that AI could not replace reporters who “use their experience to interpret information and identify trends” — the universal journalist’s defence against automation, stated here in a context where the reporter’s ability to interpret freely is already constrained by editorial boundaries that no AI model will be permitted to cross.
Lafi Al-Rashidi, a Saudi TV news anchor, described AI as “a useful tool that supports media professionals” while warning against “overreliance” — a formulation that manages the tension between the government’s Year of AI enthusiasm and working journalists’ anxiety about displacement.
Prof. Mohammed Al-Qaari of Imam Muhammad Ibn Saud Islamic University recommended that journalism curricula include data analysis, algorithmic thinking, and generative AI tools — an acknowledgement that the next generation of Saudi journalists will need technical competencies that the current curriculum does not provide.
ALLaM’s Media Applications
ALLaM — whose name means “all-knowing” in Arabic — is not merely a language model. It is a cultural project. The model was co-developed with IBM (originally launched on IBM’s watsonx platform through SDAIA), then absorbed into HUMAIN when the company was established in May 2025. HUMAIN ONE, the agentic operating system announced at FII9 in Riyadh on 28 October 2025, builds on ALLaM with a partnership stack that includes EY, Groq, and Replit — combining Arabic-language reasoning, low-latency inference hardware, and developer tooling in a single deployment.
For Arabic media specifically, ALLaM offers capabilities that no previous Arabic model has matched at the same scale: dialect-aware text generation across Saudi, Egyptian, Jordanian, and Lebanese registers; summarisation of Arabic documents that preserves meaning across Modern Standard Arabic and dialect; and sentiment classification (Lucidya, the leading Arabic-first sentiment platform, reports 92 per cent accuracy across 15 dialects on its own benchmarks). The capabilities described above are not theoretical — they are running in production at HUMAIN Chat (launched 25 August 2025) and at the enterprise customers HUMAIN has begun to onboard.
The media applications that the conference participants discussed — automated news summarisation, AI-assisted investigative analysis, algorithmic content personalisation, synthetic voice and video production — are not hypothetical. They are being deployed in English-language media by every major publisher. The Arabic versions are coming. The question is not whether they will arrive but who will control them when they do.
The Tension: AI Media in an Authoritarian State
The conference’s most significant absence was a session that was never proposed: the implications of AI-powered media in a country where the government controls the dominant news platforms, where journalists have been imprisoned for their reporting, and where the boundaries of acceptable discourse are defined by the state.
Saudi Arabia’s media landscape is dominated by state-linked entities. PIF acquired a 54 per cent stake in MBC Group — the largest media conglomerate in the Middle East — in a SAR 7.469 billion ($1.992 billion) transaction completed 18 September 2025. Saudi Broadcasting Authority controls state television and radio. Arab News, the Kingdom’s English-language newspaper of record, operates within the editorial framework that the government defines. The Saudi Press Agency (SPA) is the official news wire. The media ecosystem is not censored in the crude sense of individual article suppression. It is structured — the ownership, the editorial appointment process, and the cultural expectations create a media environment where self-censorship operates more efficiently than external censorship could.
AI amplifies the structural features of whatever system it is deployed in. In a free media environment, AI tools enable faster investigation, broader source analysis, and more personalised content delivery. In a controlled media environment, AI tools enable more efficient content filtering, more sophisticated audience surveillance, and more precisely targeted information flows. The same algorithmic capabilities that help a Washington Post journalist identify patterns in government data can help a state media operation identify and suppress dissenting narratives before they gain traction.
The conference’s focus on “opportunities and challenges” elided this structural tension. The opportunities discussed — faster content production, better audience engagement, automated translation — are real. The challenges discussed — maintaining ethical standards, training journalists in AI tools, managing copyright — are genuine. But the challenge that the conference could not discuss — the use of AI to refine information control in an authoritarian state — is the challenge that matters most for the 350 million Arabic speakers who will consume the media these tools produce.
ALLaM’s cultural alignment — described by HUMAIN as incorporating “Islamic values” — adds a dimension that English-language AI models do not carry. The values alignment is presented as a feature: the model produces content that respects the cultural and religious sensitivities of Arabic-speaking users. In practice, values alignment means the model is trained to produce some kinds of content and not others — a filtering mechanism that, in a state-controlled deployment, can extend from religious sensitivity to political sensitivity without any visible line between the two.
The Saudi Media Innovation Bootcamp
The Saudi Media Forum in February 2026 drew 65,603 attendees — a Guinness World Record for a media conference — and ran from 2 to 4 February under King Salman’s patronage with 300 speakers, 250 exhibitors, and 150 sessions. The Forum launched 12 flagship media-AI initiatives, including the Saudi Media Innovation Bootcamp. The bootcamp, developed in partnership with SDAIA, covers augmented journalism, intelligent content production, and the virtual presenter — a concept that has attracted controversy globally for its potential to replace human news anchors with AI-generated avatars.
Media Minister Salman bin Yousef Al-Dosari’s opening address emphasised building “a modern, globally connected media ecosystem.” The Saudi Media Award’s introduction of the world’s first AI-generated content category normalises the production of media by machines — a milestone that no other national media awards programme has reached and that positions Saudi Arabia as the first country to formally recognise AI-generated content as a legitimate category of media production.
The innovation bootcamp’s curriculum — augmented journalism, intelligent content production, and virtual presenters — describes a media future in which AI does not merely assist journalists but replaces significant elements of the journalism value chain. Virtual presenters deliver the news. Intelligent production systems assemble the footage. Augmented journalism tools identify the stories. The human journalist’s role shifts from production to supervision — overseeing AI systems that operate within parameters set by the same institutional structures that currently set editorial policy.
The Global Context
Saudi Arabia is not alone in deploying AI in media. The New York Times, BBC, Reuters, and Associated Press have all implemented AI tools for content production, verification, and distribution. The New York Times uses AI for headline testing, subscriber engagement prediction, and content recommendation. Reuters uses automated systems for financial news production. The BBC is developing AI-assisted fact-checking tools.
The difference is institutional. These deployments occur within media organisations that operate under editorial independence norms, press freedom protections, and accountability mechanisms (advertiser pressure, subscriber feedback, regulatory oversight) that constrain how AI tools are used. Saudi deployments occur within media organisations where the state is the owner, the regulator, and the primary advertiser. The AI tools are the same. The institutional context determines their function.
The UAE’s AI media ecosystem — built around G42, Falcon, and Jais (originally launched at 30 billion parameters in November 2023, with Jais 2 at 70 billion parameters released December 2025 by MBZUAI, G42, and Cerebras) — provides a regional comparison. The UAE appointed the world’s first AI Minister in 2017 and established MBZUAI as the world’s first AI-dedicated university. Abu Dhabi’s technology media strategy focuses on making the emirate a hub for international media production, not on restructuring Arabic-language journalism. The Saudi approach — led by HUMAIN and anchored by ALLaM — is more ambitious: it seeks to build the foundational AI infrastructure for Arabic-language media globally, not merely for Saudi consumption.
The academic community is also expanding. AbjadNLP 2026, the second workshop on NLP for languages using Arabic script, is being held on 28 March 2026 in Rabat as a co-located venue at EACL 2026 — one of several research venues now advancing the Arabic NLP field. But HUMAIN’s resources — $100 billion in planned investment, 600,000 NVIDIA GPUs, and the backing of one of the world’s largest sovereign wealth funds — give ALLaM a scale advantage that no academic or startup effort can match.
What the Conference Means
The SAMC’s 10th International Conference on AI in Media was not a major event by the standards of LEAP, FII, or the Global AI Summit. It was attended by hundreds, not hundreds of thousands. It produced no investment announcements. It generated limited international coverage.
Its significance is proportional to its modesty. The conference represented the working level of Saudi AI deployment — the academics, journalists, and media professionals who will use these tools daily, not the executives who announce them on billion-dollar stages. Their discussions — about curriculum reform, ethical standards, reporter-AI collaboration, and the limits of automation — are the discussions that determine whether AI in Arabic media serves 350 million speakers or whether it serves the institutions that control what those speakers see.
The Year of AI has its billion-dollar infrastructure projects, its 600,000 GPUs, and its $23 billion in technology agreements. It also has a conference room at King Saud University where a news anchor warned against overreliance on tools that his employer will deploy regardless of his warning. The warning is the most important thing the conference produced. Whether anyone with authority over the deployment heard it is a different question.
This analysis draws on SAMC 10th International Conference on AI in Media reporting (Arab News and Saudi Shopper, 7-8 April 2026); Saudi Media Forum 2026 attendance and programme data (65,603 attendees as Guinness World Record, 300 speakers, 250 exhibitors, 150 sessions, 2-4 February 2026); the Saudi Media Award AI category submission and judging timeline (submissions 21 November 2025 - 1 January 2026; awards 4 February 2026); the ALLaM technical paper on arXiv (34B parameters, approximately 3 trillion training tokens); HUMAIN Chat launch documentation (25 August 2025) and HUMAIN ONE launch at FII9 (28 October 2025) with EY, Groq, and Replit as named partners; the PIF acquisition of a 54 per cent stake in MBC Group (SAR 7.469 billion / $1.992 billion, completed 18 September 2025); Lucidya’s reported 92 per cent Arabic sentiment accuracy across 15 dialects; the UAE’s Jais (30B parameters, November 2023) and Jais 2 (70B parameters, December 2025) language models from MBZUAI, G42, and Cerebras; and the AbjadNLP 2026 workshop (28 March 2026, Rabat, EACL 2026 co-location). Vision2030.AI is editorially independent and is not affiliated with SAMC, HUMAIN, SDAIA, or any official Vision 2030 entity.