What the Election Commission has proposed
The Election Commission of Pakistan (ECP) has circulated 12 draft amendments to its 2023 codes of conduct to regulate artificial intelligence in elections. Minute Mirror reported on 25 October 2026 that the changes cover political parties and candidates, media organisations and election observers. The commission has invited feedback until 20 November.
According to that report, the draft would:
- prohibit deceptive deepfakes of political leaders, candidates, election officials, members of the judiciary, armed forces personnel and other state officials when the material is false, defamatory or intended to ridicule them;
- require clear labels on AI-generated or manipulated images, video and audio in campaign material;
- require disclosure at the start of each AI robocall or voice-cloned message;
- bar parties and candidates from using bots or automated accounts to manufacture artificial trends;
- require parties and candidates to preserve AI-generated material produced during the election process;
- require media organisations to disclose content created or modified with AI, and stop observers from using AI to generate or alter their findings.
The draft is not final. Reporting also describes feedback being sought on proposals rather than finished rules, and neither report mentions penalty provisions. One caveat on our sourcing: we could not reach the ECP's own site, so this analysis rests on press accounts of the draft, not its text. The Minute Mirror piece itself relies on unnamed sources and carries an AI-summary label.
The strongest case for the rules
The case for acting is real. A convincing fabricated audio clip of a candidate, released the night before polling, can spread faster than any correction. Voice cloning makes this cheap, and robocalls reach voters who never see a fact-check. The 2023 codes did not anticipate any of this. A regulator that waits for a scandal before writing rules will write them in a panic. Disclosure duties are also among the lightest tools available: they leave speech in place and only require honesty about how it was made.
The ECP also has a constitutional footing for acting. Article 218(3) of Pakistan's Constitution makes it the Commission's duty to ensure elections are conducted "honestly, justly, fairly and in accordance with law, and that corrupt practices are guarded against." Deceiving voters about whether a statement was actually made fits within that duty.
Where the draft is strong
Several choices reflect a proportionate approach. Labelling and disclosure are preferred over blanket bans on synthetic media, which would be unworkable when parties use AI for translation, captioning and design. The bot provision targets manufactured engagement rather than anyone's opinion. Record-keeping gives the Commission an audit trail without requiring pre-clearance of content. Pre-publication approval is where election content rules most often turn into censorship, and nothing reported suggests the draft goes there.
Where it needs tightening
The weakest part is the deepfake prohibition as reported. A ban on material that is false and defamatory is defensible. A ban on material "intended to ridicule" state officials is much broader. Political satire is ridicule by design, and a clearly parodic AI video of a minister is not deception because no reasonable viewer would believe it. If the final text lets ridicule alone trigger a ban, it will punish the critics the rule should protect. Article 19 of the Constitution guarantees freedom of speech and expression subject to "reasonable restrictions imposed by law", and a restriction defined by whether an official feels mocked is hard to call reasonable.
The protected class is also unusually wide. Judges, military personnel and "other state officials" are not candidates. Extending special protection to the military and judiciary in an election code goes beyond protecting voters from deception and moves toward protecting institutions from criticism. The final text should limit the prohibition to material that a reasonable viewer would take as authentic and that is likely to mislead voters about an election.
Three further gaps stand out:
- Enforcement is unsettled. The Minute Mirror report says PEMRA, a proposed Social Media Protection and Regulatory Authority and the National Cyber Crime Investigation Agency would monitor undisclosed AI content and report violations to the ECP, and that media could be directed to remove or label material. Removal directions without a defined appeal route and time limit invite arbitrary use during a short campaign window.
- Penalties are undefined. Rules without stated consequences create uncertainty. They can be ignored by the well-resourced or enforced selectively, and uncertain liability makes cautious publishers over-censor. The final code should set graduated, proportionate sanctions, starting with correction and labelling before any removal order.
- Labels need a standard. "Clear" labelling needs a minimum format, such as visible text on video and a spoken notice on audio. Without it, compliance is a guessing game for small campaigns without lawyers.
What to do before 20 November
Stakeholders should use the comment window for specific drafting fixes, not general objections. The most valuable are: a definition of deceptive that requires a likely effect on voters' beliefs; an explicit exemption for satire, parody and commentary that is identifiable as such; a narrower covered class, or at least a reason for each category; a stated appeals process for takedown directions; and a safe harbour for good-faith labelling mistakes.
Pakistan is not wrong to address AI in elections, and a disclosure-first approach is the right instinct. The test is whether the final text treats deception as the harm or treats the mere presence of mockery, or of AI, as the harm. The first protects voters, and the second protects those in power from scrutiny.