If a video of a politician saying something shocking suddenly appears on your WhatsApp group, your first question should no longer be only whether you agree with it. It should also be whether the video is real at all. Artificial intelligence can now copy a person’s face, voice, and style so well that a fake clip can look completely genuine on a small phone screen. That is exactly the problem the Election Commission of Pakistan, usually called the ECP, is trying to deal with.
In October 2026, the ECP proposed new draft rules for how artificial intelligence, or AI, may be used during elections in Pakistan. The draft changes would apply to political parties and candidates, media organisations, and election observers. The goal is simple to understand: voters should know when election content was made or changed by AI, and deceptive deepfakes should not be allowed to mislead people.
But there is one point you must understand before anything else. These are draft proposals, not law yet. The ECP has shared the draft and asked for feedback. Nothing in the reporting about this draft says that the rules are already in force, what penalties would apply, or when they would start. So this article will not tell you that something is now banned in law. It will tell you what has been proposed, what a proper AI label should look like, and how you can spot suspicious election content yourself.
What the ECP has actually proposed
The ECP has proposed amendments to the codes of conduct that were set in 2023. Those codes already cover political parties, media organisations, and election observers. The new draft adds AI-specific duties to those codes.
In plain words, the draft says that AI is not being pushed out of elections completely. A party can still use AI tools. A media outlet can still use AI to prepare, translate, present, or analyse content. What the draft wants to control is deception. If AI was used to create or change election content, the audience should be told. If AI was used to create a fake image, voice, or video that falsely shows what a person said or did, that is a different matter, and the draft proposes to prohibit it.
The ECP has linked this work to its duty to ensure honest, just, and fair elections. The concern is manipulation: a misleading AI video shared shortly before voting could change how people think about a candidate before anyone can check the facts. A clear label gives voters a chance to pause and verify.
The draft was reported by Dawn on 24 October 2026 and by The Express Tribune on 25 October 2026. The ECP has sought feedback from the public and stakeholders by 20 November 2026, and a draft can still change before any final version is issued.
The 12 versus 13 question, explained honestly
You may see different numbers in different reports, and it is worth explaining why. The Express Tribune headline describes 12 amendments. However, the same report breaks the proposals down into 5 amendments for media, 5 for political parties and candidates, and 3 for election observers. Those three groups add up to 13, not 12.
This article does not try to hide that mismatch, and it does not invent an extra amendment to force the numbers to match. We also checked Dawn’s version of the story. Dawn describes the proposed rules in detail, but it does not state a total number of amendments at all. So Dawn’s reporting neither confirms 12 nor confirms 13.
What does that mean for you as a reader? It means you should not treat 12 amendments as a settled, exact count until the ECP publishes a final version or a clearer breakdown. The substance of the proposals matters more than the headline number. The duties described below, labelling, a ban on deceptive deepfakes, rules for bots and robocalls, and duties for observers, are the parts that affect what you will see on your screen.
What political parties and candidates would have to do
Under the draft, political parties and contesting candidates would have to clearly label all AI-generated or manipulated campaign material. That includes images, videos, and audio. A campaign poster made with AI, a campaign video edited with AI, or an audio message created with a synthetic voice would all need a clear AI label.
The draft also proposes to prohibit content that falsely portrays a person’s image, words, or actions. In other words, AI cannot be used to put words in someone’s mouth or to show them doing something they never did, and then present that as real.
Deceptive deepfakes face a stricter proposed ban. The draft prohibits deceptive deepfakes of leaders, candidates, election officials, judges, armed forces personnel, and other state officials where the content is false, defamatory, or intended to ridicule. The key words here are false, defamatory, and intended to ridicule. A clearly labelled piece of satire is a different thing from a fake video designed to make voters believe a leader said something they never said. The draft is aimed at the second kind of content.
Parties would also have to preserve and retain AI-generated campaign content. Dawn’s reporting adds that parties would retain logs of AI-assisted campaign content and submit them to the ECP on request, creating a record against which a disputed video can be checked.
What media outlets would have to do
Media organisations carry a special responsibility, because people trust news outlets more than random forwarded clips. Under the proposals, media outlets must clearly disclose when election content was generated or modified through AI.
The draft does not ban media from using AI altogether. AI may still be used for preparing, translating, presenting, or analysing content, but the outlet remains responsible for accuracy and authenticity — using AI is not an excuse for getting facts wrong.
Dawn’s reporting also notes that print, electronic, and digital media outlets, as well as social media influencers, would have to explicitly label AI-altered election content. That matters in Pakistan today, because many voters get election news from influencers and social media pages, not only television and newspapers. If misleading or undisclosed content is found, the draft proposes it must be removed or correctly re-labelled when directed by a regulator or by the ECP — deletion is not always the answer; sometimes an honest label is the correct fix.
Bots, robocalls, and voice clones
Two other parts of the draft deal with tricks that are harder to see than a fake video.
First, AI bots and automated accounts would be barred from boosting engagement or manufacturing trends. That means a party could not use a network of fake accounts to make a hashtag trend, inflate likes, comments, shares, or followers, and create a false impression of public support. If you see a sudden flood of identical comments under a political post, or a hashtag that appears everywhere within minutes with very similar wording, that is the kind of activity the draft aims to stop.
Second, AI robocalls and voice-cloned voter-outreach messages would have to disclose at the start that they are AI-generated. This is a very practical rule. Imagine receiving a phone call in a familiar politician’s voice, asking you to vote a certain way. If that voice was cloned by AI, you deserve to know that in the first few seconds, not after you have already been convinced. The draft says that disclosure must come at the start of the call or message, where it cannot be missed.
What election observers would have to do
Election observers are the independent eyes and ears during an election. They watch polling, note problems, and report what they saw. The draft gives them two clear duties.
First, observers must stay vigilant for AI manipulation. If they suspect a deepfake or a piece of disinformation, they must report it to the ECP promptly, without spreading it further. That last part is crucial. Forwarding a suspected fake just to warn people often spreads it faster than the original poster ever could. The draft’s approach is simple: report it, do not amplify it.
Second, observers must not use AI to generate, alter, or enhance the material they submit to the ECP. That includes observations, findings, photographs, videos, and reports. An observer’s report must reflect what actually happened, not an AI-improved version of it. If a photograph is enhanced to make a crowd look bigger, or a report is rewritten by AI in a way that changes the meaning, the ECP cannot trust it. Accuracy in observer reports protects the credibility of the whole election.
Who would enforce the draft rules
As proposed, enforcement would involve several bodies, not only the ECP. The Pakistan Electronic Media Regulatory Authority, known as PEMRA, the Social Media Protection and Regulatory Authority, and the National Cyber Crime Investigation Agency, known as the NCCIA, would monitor undisclosed or unlabelled AI media content and report violations to the ECP.
The ECP, in turn, could order takedown or re-labelling of the content. Dawn’s reporting describes the ECP as retaining authority to order immediate content takedowns or re-labelling, based on compliance reports from those monitoring bodies.
Again, remember the status of all this. This is the enforcement model in the draft. No penalties, no appeal process, and no start date are stated in the reporting found so far. Readers should therefore be careful about any social media post that claims a specific punishment already applies. Until a final version is published, those claims are speculation, not fact.
How to spot AI election content: a practical checklist
You do not need special software to catch many AI fakes. You need a few seconds of patience and a habit of checking before sharing. Work through this checklist whenever a political clip feels surprising, emotional, or too perfect.
First, look for a label. A compliant post should say clearly that it was made or altered with AI, in a place that is easy to see. No label on a polished, dramatic video is a warning sign and a reason to verify, though not proof of fakery on its own.
Second, watch the face and mouth. Lip movements may not match the words, blinking may look unnatural, and skin may look too smooth, almost like plastic. Hands are another weak point: fingers may look merged or bent in odd ways.
Third, listen to the voice. A cloned voice may sound flat, with odd pauses or missing breathing sounds, and background noise may vanish when the person starts speaking, suggesting the audio was replaced.
Fourth, check the background. Text on signs or clothing may look garbled or misspelled, objects may melt into each other, and shadows may fall in impossible directions.
Fifth, ask whether the claim makes sense. Would a senior leader really announce a major decision only in a shaky forwarded WhatsApp video, with no party account or news outlet reporting it? If the only source is a forwarded message, be extra careful.
Sixth, search before you share. Check a distinctive phrase, a reputable news outlet, and the politician’s or party’s official account. If no credible source confirms it, do not forward it as fact.
Seventh, watch for emotional manipulation. Deepfakes are designed to make you angry, shocked, or afraid, because strong emotions make people share without thinking. If a clip makes you want to forward it immediately, pause for thirty seconds and run the checks above.
What a proper AI label looks like
A good label is not a puzzle. It should answer three questions at a glance: Was AI used? What was changed? And who is responsible for the content?
A clear label might say, for example, that an image was generated with AI, or that a video’s voice was created with AI. It should appear on the content itself or directly beside it, not buried in a comments section or a separate page that most viewers will never open. A label that only appears after you click through three menus is not a real label in practice, even if it technically exists somewhere.
For robocalls and voice messages, the draft sets a stricter standard: the disclosure must come at the start. That makes sense. A listener who hears a familiar voice for two minutes before learning it was synthetic has already been influenced. The opening seconds are where the warning belongs.
For news content, a label should also sit alongside the outlet’s name. When a trusted outlet labels its own AI-altered content, it protects its own credibility. When an outlet quietly uses AI without telling viewers, and the audience later discovers it, trust is damaged far beyond that one story.
| Who the draft covers | What must be labelled or disclosed | What the draft proposes to prohibit |
|---|---|---|
| Political parties and candidates | All AI-generated or manipulated campaign images, videos, and audio | Content falsely showing a person’s image, words, or actions; deceptive deepfakes of leaders, candidates, officials, judges, armed forces personnel, and other state officials where false, defamatory, or meant to ridicule |
| Media outlets and influencers | Any election content generated or modified through AI | Publishing misleading or undisclosed AI content without removing or correctly re-labelling it when directed |
| AI bots and automated accounts | Nothing to label, because this activity would be barred | Using bots to inflate likes, comments, shares, or followers and manufacture false trends |
| AI robocalls and voice clones | AI nature disclosed at the very start of the call or message | Sending voice-cloned voter messages without that opening disclosure |
| Election observers | Suspected deepfakes reported promptly to the ECP | Using AI to generate, alter, or enhance observations, photos, videos, or reports sent to the ECP |
What to do when you see suspicious election content
One forward from a trusted family member carries more weight than a hundred posts from strangers, so follow a safe routine.
Do not forward content until you have checked it. If unsure, say so to the sender rather than passing it on with is this true? — that still spreads the clip. Check the original source: a real major statement is rarely reported by only one anonymous page.
If AI content is clearly labelled, share it with the label intact. Cropping out a label or re-uploading a video without its caption turns honest content into misleading content. If something looks like a deliberate fake, use the platform’s report option for false or manipulated media, and watch for official ECP announcements about reporting once final rules exist. In group chats, mute groups that keep sharing unverified videos, ask for sources before reacting, and do not be the first to forward everything.
What the draft does NOT do
It is just as important to understand what this draft does not do, because misunderstandings spread as fast as deepfakes.
First, it does not ban AI from elections. Media may still use AI for preparing, translating, presenting, and analysing content, with accuracy remaining their responsibility, and parties may still use AI in campaigns if the material is labelled and honest.
Second, it does not create a final law today. It is a proposal open for feedback until 20 November 2026, so anyone claiming a particular penalty already applies is going beyond what the reporting says.
Third, it does not name a specific election. The proposals are framed for elections generally; do not attach them to one particular upcoming election date unless the ECP itself does so later.
Fourth, it does not make every AI-labelled video illegal, nor every unlabeled video automatically fake. A label is about transparency, and the only way to know the truth about a specific clip is to verify it through credible sources.
Fifth, it does not solve the problem on its own. Labels, monitoring, and takedown orders help, but the strongest protection remains an audience that pauses, checks, and refuses to spread unverified claims.
Feedback is open until 20 November 2026
If you are a voter, a journalist, a party worker, or an observer, this is the window in which your view can count. The ECP has sought feedback from the public and stakeholders by 20 November 2026. Because this is a draft, the final rules may look different from the proposals described here.
You do not need to be a legal expert to have a useful opinion. Practical questions matter most: Are the labelling rules clear enough for a small social media page to follow? How quickly should a disputed video be taken down or re-labelled? What happens to ordinary users who unknowingly forward a fake? Raise such points through the official channels the ECP announces for this consultation, rather than through social media rumours. Until a final version is published, check the source and do not repeat a detail you cannot verify.
Why this matters for ordinary voters in Pakistan
Pakistan’s voters increasingly get their news from phones, not only television. A video can reach millions within hours through WhatsApp, TikTok, Facebook, and YouTube — and corrections travel far more slowly than the original lie.
A clear labelling rule changes that balance. A labelled party video can be enjoyed as creativity, not mistaken for real footage. A labelled news report lets viewers separate reporting from illustration. A robocall that announces itself as AI lets listeners judge how much weight to give it. But labels only help if people read them, and monitoring only works if suspicious content is reported rather than forwarded.
The safest habit you can build today does not depend on whether these draft rules become final. Pause before you share. Look for a label. Check a credible source. And if something feels designed to make you angry in seconds, give it a minute of calm checking first. That minute is often all it takes to stop a fake in its tracks.
For the full reporting on these draft proposals, see The Express Tribune, which reported the proposed amendments, the labelling duties for parties and media, the proposed monitoring role of PEMRA, the Social Media Protection and Regulatory Authority and the NCCIA, and the 20 November 2026 feedback deadline.
FAQs
Is the ECP’s AI election rule already law in Pakistan?
No. As reported in October 2026, these are draft amendments to the 2023 codes of conduct, shared for public and stakeholder feedback by 20 November 2026. No commencement date, penalties, or appeal process were stated in the reporting found, so do not treat any part of the draft as already in force.
What must be labelled under the draft proposals?
Political parties and candidates would have to clearly label all AI-generated or manipulated campaign images, videos, and audio. Media outlets would have to disclose election content generated or modified through AI. AI robocalls and voice-cloned voter messages would have to state at the very start that they are AI-generated.
How can I tell if an election video is a deepfake?
Look for a clear AI label first, then check lip movement, unnatural blinking, odd hands, garbled background text, and a voice that sounds flat or lacks natural breathing. Most importantly, verify the claim through the person’s official account or a reputable news outlet before sharing, and do not forward a clip that you cannot confirm.
What should I do if I receive a suspected fake election video?
Do not forward it, even to ask whether it is true, because forwarding spreads it further. Check the original source and credible news reports, keep any label intact if you share verified content, and use the platform’s report option for manipulated media. Election observers are specifically asked in the draft to report suspected deepfakes to the ECP promptly without spreading them.
Also read
Pakistan’s Buraq Urdu AI Model: What Was Announced in Seoul — and What Is Not Ready Yet