The Electoral Commission’s published findings describe what its pilot detected and where it saw risks. The distinction between that observed evidence and wider claims about AI election interference matters.
Election deepfake findings and the measured items
The Commission reports 219 political deepfakes identified after automated processing of around three million posts and human review of a subset. It says 15 posts met its threshold for electoral-integrity risk, including false candidate activity or announcements and misleading polling-day information. These are reported observations within the exercise, not proof that every misleading item used AI or influenced a voter.
https://www.electoralcommission.org.uk/cy/node/33666
Absence of detected interference has a defined meaning
The account says it detected no evidence of deliberate use of highly convincing political deepfakes to target or interfere with the May 2026 elections. It also warns that this does not mean future risks are absent. A finding from particular monitoring should not be converted into a claim that all synthetic media is harmless or that interference can never succeed.
https://www.electoralcommission.org.uk/cy/node/33666
Keep survey responses and monitoring separate
The same findings discuss voters reporting that they had seen misinformation or deepfakes. Those responses concern people’s perceptions and experiences; they are different from verified items identified by the monitoring tools. The launch document also gives the pilot’s scope in England, Scotland and Wales, excluding Northern Ireland in that phase, which limits any claim about full UK coverage.
https://www.electoralcommission.org.uk/media-centre/electoral-commission-launches-deepfake-detection-pilot-counter-ai-misinformation
Sources and further reading
https://www.electoralcommission.org.uk/media-centre/electoral-commission-launches-deepfake-detection-pilot-counter-ai-misinformation
https://www.electoralcommission.org.uk/cy/node/33666