The Scale of the Problem
On 30 May 2026, Demos published polling conducted by Opinium between 30 April and 6 May — the month before the UK's 7 May local and devolved elections. The headline finding was striking: 30% of UK adults, approximately 16.5 million people, reported encountering political deepfakes online in that period. Among them, 16% saw such content more than five times. The three most-deepfaked politicians were Donald Trump (seen by 45% of those exposed), Keir Starmer (36%), and Nigel Farage (27%). More than half of the deepfakes encountered were negatively framed — content designed, in most cases, to damage rather than persuade.
The distribution channels were not fringe platforms. TikTok accounted for the highest reported exposure (35%), followed by Facebook (34%) and Instagram (26%). These are services already subject to the Online Safety Act 2023 and, for their EU-facing operations, the Digital Services Act. The problem is not invisible. What is absent is any mechanism compelling platforms to remove, label, or flag synthetic political content during an election period.
What the Bill Does — and Doesn't — Do
The Representation of the People Bill is the government's current vehicle for electoral reform. Committee stage ran across nine sittings from 18 March to 16 April 2026, covering campaign finance, voter registration, and transparency requirements for digital political advertising. What the bill does not contain is any duty requiring platforms to take down AI-generated electoral deepfakes, any labelling mandate for synthetic political content, or any rapid-response enforcement mechanism during regulated periods.
Cross-party amendments to fill that gap were tabled at committee stage and rejected. Demos had proposed provisions creating clear legal responsibilities for platforms and developers. Full Fact called for making deepfake impersonation of candidates a specific criminal offence. Multiple MPs argued for disclosure rules: Emily Darlington pushed for updates to electoral law covering disclosure of AI-generated digital content; Rushanara Ali warned that hostile actors were already "exploiting online platforms to flood the debate with disinformation and deepfakes." The government said it was sympathetic to the intent behind the amendments but rejected all of them, committing to a fuller response to the Rycroft Review "in due course."
Chief Science, Innovation and Technology Committee chair Chi Onwurah described the situation as "deeply alarming." James Cleverly, for the Conservatives, supported proportionate labelling measures in principle but counselled careful consultation rather than rushed drafting. The result is a bill that modernises election administration while leaving AI-powered electoral manipulation entirely unaddressed.
The Government's Case — Taken Seriously
The case for deferring is not absurd, and it deserves a fair hearing before being dismissed. The government points to the Online Safety Act 2023, under which foreign interference is designated a priority offence requiring platforms to proactively prevent exposure to such content. Ofcom's enforcement powers — fines up to £18 million or 10% of global revenue — are substantial, and further duties expected to come into force in 2027 will require platforms to increase transparency on content removal.
The Rycroft Review, published 25 March 2026, recommended a National Counter Disinformation Centre modelled on bodies in France and Sweden, and proposed stronger Electoral Commission powers to demand real-time platform data during election periods. A DSIT consultation on "digital replicas" is expected in summer 2026 — a broader exercise that would include political deepfakes. The consultation-first argument has genuine merit where legislation risks overreach: a poorly drafted platform liability rule for "AI-generated political content" could capture legitimate satire, news editing, and campaign advertising without clear definitions.
Why the Existing Framework Does Not Reach the Harm
The problem is structural. The Online Safety Act's election provisions are narrow by design. Political deepfakes are only actionable under the Act where they can be proved to constitute a foreign interference offence — which requires demonstrating state-backed origins. That is an evidentiary bar that almost no piece of viral electoral content will clear, particularly when coordinated amplification is designed to obscure its source. Full Fact's analysis of the Act concludes it "does not extend to most harms from generative AI misinformation" in the electoral context.
The Electoral Commission launched a deepfake detection pilot ahead of the May elections — a genuine operational step. But its own timeline acknowledges results will not arrive for at least six months. That is a detection programme for 2027, not for the elections that 16.5 million voters just navigated without protection.
The EU's posture offers an instructive contrast. Under the Digital Services Act, platforms with more than 45 million EU users must proactively assess and mitigate systemic risks to electoral processes, and must remove illegal deepfakes upon notification. The Code of Practice on Disinformation was formally integrated into the DSA framework in February 2025, making deepfake labelling a compliance benchmark rather than a voluntary pledge. The EU AI Act adds a further transparency requirement for synthetic media. None of that applies in the UK, and the Representation of the People Bill does not move the UK any closer to equivalent protections.
What Proportionate Reform Looks Like
The government is right that durable legislation demands careful definition. It is wrong to treat that as a reason for unlimited deferral. The digital imprint rules already in the Bill — requiring disclosure of who paid for digital political advertising — could be extended to AI-generated synthetic media with modest additional drafting. A mandatory labelling requirement for election-period content on designated platforms is not a novel concept; the EU has already shown it can be implemented without suppressing legitimate commentary. A rapid-notification mechanism for candidate impersonation deepfakes, with a contestation window, is achievable within the bill's existing enforcement architecture. And the Rycroft Review's own recommendation — real-time Electoral Commission powers during election periods — could be added without reopening the entire legislative framework.
None of these require resolving every definitional edge case. They require political will.
The Shrinking Window
Sixteen and a half million UK voters encountered AI-generated political manipulation in the month before the May 2026 elections. Forty-two percent of the population expressed concern about deepfakes affecting the vote. The government acknowledged the threat, rejected the fixes offered, and promised consultation. The Representation of the People Bill is still progressing through Parliament — the window to amend it has not closed. The alternative is arriving at the 2027 general election with the same regulatory gap, and deepfake technology that has had another year to improve.