The Clip Shows the Nominee Endorsing Them. The Nominee Was Never in the Room.
Deepfake endorsements and synthetic testimonials are seeding party-list recruitment for 2028, and a first-time voter has no legal tool to confirm the words came from a real mouth.
A party-list group DMs you a 40-second clip. A recognizable public figure looks into the camera, names the group, and tells you to vote for it in 2028. The lighting is clean, the lip-sync is close, the caption says "official endorsement." Nobody in that clip has to have said any of it.
This is the recruitment layer, not the campaign proper, which is exactly why it slides past scrutiny. Party-list groups are courting members, sectoral leaders, and barangay organizers well ahead of the ban period, and synthetic video is doing the persuading before a single official ad runs.
The endorsement nobody has to give
The tools are cheap and the output is good enough for a phone screen at arm's length. Face-swap and voice-clone models can now stitch a plausible endorsement from a few minutes of a person's existing footage, and public figures leave plenty of that lying around. The "grassroots" testimonials are worse, because the faces are invented outright, and an invented face cannot file a complaint.
What travels with the clip is a story: a farmer in the province, a tricycle driver, a nursing graduate, all vouching for a group you have never heard of. None of them exist. The comment section fills with accounts that also do not exist, and the whole thing reads like your barkada finally getting political.
Comelec regulates the campaign, not the seeding
Comelec has issued rules on artificial intelligence and disinformation in election material, requiring disclosure and letting the poll body order takedowns. Those rules bite hardest on registered candidates during the campaign period, on paid ads, on accredited pages. A recruitment reel forwarded through Messenger during an off-cycle year sits in the blind spot, and the poll body would need to know the clip exists before it can act on it.
Enforcement also assumes you can trace the sender. That is where the second law fails you.
SIM registration was sold as the trace. It isn't one.
The SIM Registration Act was pitched as the fix for scam texts and anonymous accounts, tying every number to a verified identity. The seeding does not run on your registered SIM. It runs on Facebook Reels, TikTok, Telegram channels, and pages managed through accounts that never touched a Philippine SIM, some run from content shops, some booted up with boxes of pre-registered numbers bought secondhand.
So the law gives the state a database and gives you nothing. As a first-time voter, you cannot query who uploaded the clip, whether the endorsement was authorized, or whether the person on screen ever opened their mouth. The burden of proof lands on the target of the deepfake, who has to notice, gather evidence, and chase a takedown while the clip keeps reshared.
The gap is structural. Platforms label some AI content and miss most of it, Comelec can only pursue what gets reported inside its jurisdiction, and the trace law was built for a different crime. What reaches you is a video with no chain of custody and no way to check one. You are asked to trust a clip that was designed so you cannot verify it, and the next 20 you scroll past will be built the same way.