Two Anwars on the Ballot? Regulating AI Before GE16

Malaysia’s 16^th general election (GE16) might feature two versions of Prime Minister Anwar Ibrahim. On 18 July 2026, the prime minister launched PMX AI, a digital double developed by Zetrix AI as a Parti Keadilan Rakyat (PKR) initiative. The model has been trained on his speeches, writings and policy record. Designed to look and sound like Anwar, it answers public questions through WhatsApp, helps users navigate government services and gathers voters’ sentiment. Its arrival blurs the line between public-service communication and party-political messaging in a country without election-specific rules for AI-generated content.

Recent deepfakes impersonating Anwar and Sultan Ibrahim underscore the extent of the problem. Malaysia’s short electoral timetable magnifies the risk. Campaigns typically last about two weeks and ballots are counted quickly, leaving little time to counter synthetic content. Although GE16 is constitutionally due by February 2028, speculation about a snap poll in late 2026 or 2027 suggests it could arrive before Malaysia’s emerging AI governance architecture is tested.

That urgency is sharpened by the slow progress of Malaysia’s AI framework. In November 2025, Digital Minister Gobind Singh Deo said the first AI Governance Bill was nearly complete, with a wider framework expected to reach Cabinet in June 2026. Yet a month after, on 10 July 2026, the National AI Office began public consultation on the Bill, which remains in the pre-drafting stage. More importantly, general AI legislation and election integrity rules are not the same thing. Malaysia has no publicly announced election-specific rules governing synthetic media, AI-assisted campaign material or deepfake attacks during elections.

The gap matters because entities employing AI-enabled deception can exploit Malaysia’s expanding digital infrastructure. Online scam losses reached RM2.97 billion in 2025, up from RM1.57 billion in 2024, with another RM830 million lost in the first five months of 2026. These figures do not isolate losses caused by deepfakes, but they show how synthetic trust (credibility manufactured through AI-generated voices, images and personas) can be monetised through fake identities and platform-amplified deception.

The same infrastructure can be repurposed for political purposes. During GE15 in 2022, ethnoreligious rhetoric invoking the 13 May 1969 riots spread widely on TikTok, prompting government pressure on ByteDance and the removal of more than 1,000 videos after the election. There is no evidence that generative AI drove this activity during GE15. It could, however, accelerate such multilingual, ethnoreligious and personality-driven operations. Content that once required teams working across Malay, Mandarin, Tamil and English audiences can now be produced faster through prompts, templates and platform-specific optimisation.

Malaysia should coordinate with the region. Vietnam’s AI law and Singapore’s agentic AI governance framework are not election rules, but they show how quickly AI governance is advancing across Southeast Asia.

Yet Malaysia’s laws are designed for general online harms and cybercrime, not the compressed timelines, disputed attribution, disclosure requirements and institutional coordination of an election campaign. The Online Safety Act 2025 and the Communications and Multimedia Act 1998 allow action against unlawful content after it appears. The Cybercrime Bill 2026 criminalises offences involving deepfakes, computer-related forgery and digitally manipulated intimate images, but intent, purpose and consequence must be proven. The proposed AI Governance Bill would assign clearer responsibilities to AI developers and deployers according to their control over a system across its lifecycle. None, however, provides an election rulebook. For instance, can a political party circulate an AI-generated video of its own leader? Should AI-generated political smears be labelled as such? Who decides whether a viral synthetic clip is misleading voters during the final 48 hours before polling — the Election Commission or Malaysian Communications and Multimedia Commission (MCMC)? The temporary restrictions on Grok in January 2026 showed how reactive platforms struggled to maintain control once harmful content had spread.

The Philippines offers a regional precedent. The Commission on Elections issued Resolution No. 11064 for the 2025 elections, requiring disclosure of AI use in campaign materials. The resolution sought to address deepfakes, fake accounts, bots and other digital practices. The enforcement was imperfect and civil society raised concerns about speech and registration requirements. But the experience gives Malaysia a tested template.

Policy Implications

Malaysia can act while awaiting a full AI law. Election-specific AI regulation would reinforce broader AI governance and strengthen the Election Commission’s oversight powers in tandem with MCMC. The two could jointly require disclosure of AI-generated or AI-manipulated campaign material. The aim is not to police every meme, parody or routine edit, but to prevent voters from being misled about the source, authenticity or evidentiary value of political content.

A campaign-period mechanism is also needed for volatile moments, especially the final days before polling. A public reporting channel should enable voters and parties to flag suspected deepfakes, preserve content and metadata and trigger immediate platform action or later investigation. The Election Commission, MCMC, social media platforms, civil society fact-checkers and political parties could coordinate provenance checks, emergency labelling, platform escalation and public clarification when synthetic content threatens to mislead voters.

Preventive measures require a clearer punitive framework. Malaysia should examine whether the Election Offences Act 1954 should cover malicious synthetic impersonation, including deepfakes that misrepresent opponents or falsely promote candidates. Because the Act does not expressly address AI-generated election content, it remains unclear whether such conduct can trigger electoral-misconduct proceedings or stronger penalties for producers.

Malaysia should coordinate with the region. Vietnam’s AI law and Singapore’s agentic AI governance framework are not election rules, but they show how quickly AI governance is advancing across Southeast Asia. Shared election-integrity protocols would help ASEAN states respond to common platforms, tools and cross-border influence risks without having to start from scratch at each election.

The AI Governance Bill will matter for the long term, but GE16 may arrive first. Malaysia need not come up with such a framework before the next campaign. It must regulate AI-assisted campaigning before campaigning begins.

2026/210