Malaysian Prime Minister Anwar Ibrahim will deploy an AI-generated digital avatar to engage with voters during the Johor state election this Saturday, injecting synthetic media directly into the machinery of democracy. The move, confirmed by multiple local news reports, marks one of the most high-profile uses of AI persona technology by a sitting head of government in an active electoral campaign.

A campaign bot wearing the prime minister’s face

Anwar’s team has created a realistic, AI-driven likeness of the prime minister capable of delivering scripted messages, answering voter queries, and possibly interacting in real time through natural-language processing. While exact technical details are scarce, the avatar almost certainly relies on a combination of deepfake video synthesis, voice cloning, and a large language model (LLM) to simulate the politician’s presence. Social media posts and campaign platforms will be the primary distribution channels.

The avatar is not a static recording. Early indications suggest it can be updated dynamically to reflect the latest campaign talking points, weather conditions in target constituencies, or even local dialects—something a pre-recorded video cannot easily match. This flexibility is both its political strength and its ethical hazard.

What it means for you—and for everyone who consumes political content

If you are a Malaysian voter in Johor, you may soon interact with a digital ghost of your prime minister without fully realizing it is not a real, live human being. For the rest of the world watching, this is a preview of campaigns to come.

For everyday Windows users, the technology behind Anwar’s avatar is not exotic. Windows 11 ships with Microsoft Copilot, and you can access tools like Designer, Clipchamp, and Cocreator that leverage generative AI. Third-party applications for face-swapping, voice cloning, and LLMs are a simple download away. The democratization of these tools means political operatives everywhere can create convincing AI personas with consumer-grade hardware.

For power users and IT professionals, the implications go deeper. You may be asked to help secure corporate networks against AI-generated phishing that uses synthetic video or voice. Employees could receive fake video messages from a “CEO” instructing a wire transfer, or a political deepfake could undermine a company’s reputation by spreading internally. The same avatar technology Anwar uses can be weaponized for disinformation, and Windows-based enterprise tools will need to evolve to detect it.

For developers, the rise of political AI avatars raises urgent questions about platform responsibility. If your app enables the creation of photorealistic avatars, what safeguards do you have in place to prevent misuse during elections? Microsoft’s Azure AI services, for instance, allow developers to build custom neural voices and digital humans but require customer vetting and transparency disclosures. The Anwar case tests whether such policies can withstand real-world political pressure.

How we got here: From scripted TV to synthetic candidates

The use of media manipulation in politics is old. Campaigns have long scripted candidate appearances, edited video for maximum impact, and even used body doubles for security reasons. What has changed is the cost and accessibility of producing completely synthetic, interactive personas.

Just two years ago, creating a convincing deepfake required expensive hardware, machine learning expertise, and a large data set of the target’s voice and face. Today, a campaign staffer with a mid-range Windows laptop, a subscription to an AI avatar service, and 30 minutes of source footage can produce a digital twin that passes casual inspection.

This timeline captures the acceleration:

  • 2018: Deepfake technology emerges as a niche hobby, mostly used for non-consensual pornography and internet memes.
  • 2020: The U.S. election cycle sees the first serious attempts to use deepfakes in political ads; platforms scramble to update policies.
  • 2022: Russia’s invasion of Ukraine features a deepfake of President Zelenskyy calling for surrender, though it is quickly debunked due to poor quality.
  • 2023: Generative AI tools like ChatGPT and Midjourney go mainstream, and voice-cloning services become widely available. Microsoft integrates GPT-4 into Bing and Windows.
  • 2024: India’s general election witnesses candidates using AI-generated voice calls in multiple languages, and South Korea’s presidential campaign deploys an AI avatar of a candidate to interact with voters 24/7.
  • 2025: In the Johor by-election, a sitting prime minister deploys a campaign avatar, bringing the technology into the highest level of democratic politics in Southeast Asia.

What to do now: Practical steps for navigating an AI-infused election

Assume that every video or voice message you encounter during a political campaign could be synthetic—even those from official sources. Here’s how to protect yourself and your organization:

  • Check the source. Did the content come from a verified party or campaign account? If not, treat it as suspicious. Even verified accounts can be compromised, but it is a critical first filter.
  • Look for inconsistencies. Current deepfakes often fail at realistic eye movements, unnatural blinking patterns, or distortions around the mouth and chin. Pay attention to the background; AI avatars can blur or warp environmental details.
  • Listen for oddities. Cloned voices may sound flat, lack natural breathing pauses, or mispronounce local terms. If it feels off, trust that instinct.
  • Use detection tools. Microsoft’s Video Authenticator and the Coalition for Content Provenance and Authenticity (C2PA) standard are beginning to show up in Windows media players and browsers. Real-time indicators, such as the “AI-generated” badge in Microsoft Edge, can signal synthetic content.
  • Report suspicious content. On Windows, you can use Microsoft Defender SmartScreen to flag phishing attempts that use deepfakes. Social media platforms also have reporting channels; use them aggressively during election periods.
  • For IT admins: Deploy updated endpoint security policies that flag media files with inconsistent metadata. Educate your workforce on voice phishing (vishing) and video-based impersonation risks—this is no longer science fiction.

Outlook: The ballot box meets the black box

Anwar’s AI avatar is a test balloon. If it proves effective—driving higher engagement, especially among younger voters—it will be copied. The next generation of AI avatars will be harder to distinguish from real human candidates, and real-time interactivity will make them feel even more authentic. Campaign finance laws, which were written for billboards and television spots, will be obsolete in a world where a candidate can be virtually omnipresent at near-zero marginal cost.

Microsoft and other platform companies will face immense pressure to enforce content provenance standards across Windows, Office, and Azure. The C2PA initiative, which embeds cryptographically signed metadata into images and videos, may become mandatory for politically sensitive content. But technology is only part of the answer; voters must also adjust their expectations. In an era where anything can be faked, the default assumption must shift from “I’ll believe it when I see it” to “I’ll believe it only when multiple independent sources confirm it.”

The Johor vote will be a real-world experiment in whether such caution can coexist with democratic participation. The outcome will not only affect one Malaysian state but will signal what is about to hit every Windows user’s screen.