A Nigerian startup says its AI and cryptography can make election footage hard to fake and hard to dismiss in court. Pawavox has about nine weeks before the presidential vote, a ₦13 million grant, and a pilot that produced 142 verified reports.
Pawavox signs each recording with proof of when, where and on which phone it was made, so any later edit breaks the signature. Co-founder Jason Omu says the aim is evidence that stands up when a petition reaches a tribunal. AI does part of the work, from spotting footage filmed off a screen to answering questions about field reports in four Nigerian languages. Nigeria has about 177,000 polling units, and the startup plans to train toward 5,000 observers for the 2027 elections. The company has talks with larger groups, but its only closed deal so far is the grant.
INEC’s calendar puts the presidential and National Assembly vote on 16 January 2027, with governorship polls on 6 February, according to EconoTimes. Start with the problem Omu describes. Much of the cheating, he argues, happens away from international observers, in wards and collation centres where cameras are unwelcome. When footage does survive, it faces a second test in court. Digital recordings are tested hard under Section 84 of the Evidence Act, which covers computer-produced evidence. Defence lawyers can also claim that a video was edited or made with AI.
Pawavox was built over about ten months at Fast Forward Venture Studio, with backing from Prosperity Tech. It began as a tool for collecting feedback and then turned toward verification. The app adds four proofs to a file at the moment of capture.
| Proof | What it shows |
|---|---|
| Hash. | A digital fingerprint that changes if the file is altered, even slightly. |
| Device ID. | Which phone recorded it. |
| Timestamp. | When it was recorded. |
| Location. | Where it was recorded. |
Omu says the app is also built to catch tricks before signing. It checks whether a location was faked with a VPN. If a phone is seized and starts sending false reports, its hardware-bound keys let administrators isolate that device and strip its submissions. These are the founder’s descriptions of how the system works. The only live test so far was the Osun State election.
Next comes the danger to the person holding the phone. In 2023, party agents and observers who followed ballot boxes had their phones seized and searched, according to the article behind this story. Pawavox keeps evidence in an encrypted vault, apart from the phone’s gallery. If someone demands to see the camera roll, it appears empty. In risky places, observers are told to send voice notes instead. “Their safety is more important to us than submitting data,” Omu said.
Bad networks are the other test. Signing works offline and the evidence stays on the phone. Once a weak 2G signal appears, small signed text records go first, through an open protocol called Nostr. The heavy video follows later to Google Cloud. “We built this Nigeria-first because we are Nigerians, and we know there is low connectivity,” Omu said.
Once reports arrive, election situation rooms need a way to read them. This is where most of the AI sits, and it does three jobs.
The first is checking the footage. Omu says an analysis layer looks for signs that a video was filmed from a phone or laptop screen, such as a deepfake played back to the camera. It runs before the file is signed, so a faked clip is caught before it gets a digital seal. EconoTimes reports a planned link to TrueMedia, a deepfake-detection service, so that command centres could screen clips from the public. The tool would give a likelihood flag and a confidence score for a person to review. It would not give a final ruling on whether a clip is real.
The second job is answering questions. Pawavox first let users query the data through Anthropic’s Claude. That proved too technical for civic staff, so the team built ElectionGPT, a chat tool closed off from the open web. It answers only from submitted field reports and links each answer to the videos behind it. Ask how many people wore red shirts in Lagos, and it replies with a count and the clips it came from. Ask about something outside its data, such as who won the United States midterms, and it declines. Omu says this design is what keeps it from making things up.
The third job is language. Polling units are loud, and voters speak Yoruba, Igbo, Hausa and Pidgin, often at once. Pawavox uses language data supported by YarnGPT to transcribe and translate field reports. When a transcript is unclear, the tool does not guess. It marks the words, gives a confidence score, such as 70 or 80 per cent, and asks a human analyst to listen to the clip. In each case, AI sorts and flags, and a person makes the call.
Now the scale. In Osun, Pawavox pooled 2,000 unpaid volunteers and onboarded 300 of them. They produced 142 verified reports. Many people posted updates in WhatsApp groups instead. By our sums, that is under one verified report for each onboarded user. Nigeria has about 176,974 polling units, and INEC has said it will deploy about 1.4 million ad hoc staff. Omu says opening the app to untrained citizens is off the table for 2027 because of cost. Instead, the plan has two parts.
| Plan for 2027 | Detail |
|---|---|
| Own network. | Grow toward 5,000 trained observers, about one for every 35 polling units by our sums. |
| Partners. | Pitch civil society groups to build Pawavox’s verification into the monitoring apps they already run. |
| Money. | A ₦13 million Judges’ Choice award from Yiaga Africa’s 2026 AI and Democracy sandbox, with technical support, over nine months. |
| Talks. | Technical experts at Yiaga Africa, plus IFES, NDI and Microsoft. No contract signed. |
Omu was open about the business side. “We have not signed a formal commercial contract with them beyond the grant disbursement,” he said. The Yiaga Africa award was announced on 15 October 2026. It is meant to fund infrastructure, observer coordination and archiving. By Omu’s account it will not pay to protect observers across all polling units. Beyond 2027, he sees Pawavox as a social enterprise that lives on grants first and software sales later. Kenya’s elections are the next target. Companies that need proof that a project took place where and when it should have are another.
Looking to 2031, Omu hopes to replace phones in hostile zones with discreet hardware such as smart glasses, rings or body cameras, once budgets allow.
Why it matters: Nigerians have learned that a result sheet is only as strong as the proof behind it. Portals like INEC’s IReV publish final figures. A tool that records what happened at the polling unit, in a way a court can trust, fills a different gap. The deepfake era makes that gap wider. Real footage can be waved away as fake, and fake footage can pass as real. Pawavox is also a test of whether African civic tech can find a business model between election years. The Osun numbers show how hard it is to turn interest into reports.
What to watch next: whether Pawavox signs a paid deal with Yiaga Africa, IFES, NDI or another group before the vote. Watch how many trained observers it reaches, and in which states. Watch for independent tests of its signing and anti-spoofing claims. Watch whether a tribunal ever accepts Pawavox-signed evidence. Watch how it keeps observers safe in hostile areas. And watch whether INEC or the courts say anything on the use of tools like this.
Sources: Technext interview with Jason Omu on Pawavox and ElectionGPT, 10 November 2026; EconoTimes on Yiaga Africa’s award to Pawavox’s Election GPT; Blueprint on INEC’s plans for polling units and ad hoc staff.
A note for startup founders: Don’t practice on investors. Practice on AI first. RaiseReady by Index Prima is an AI fundraising simulator for African founders: upload your pitch deck, get a readiness score and face a tough AI investor before the real meeting. Find this post, and every post we publish, at raiseready.indexprima.com.
