Ofcom has published draft proposals that would, for the first time, legally require online platforms to flag AI-generated political content with visible provenance markers, a move that could place the UK among the earliest jurisdictions to mandate labelling of synthetic election media. With local elections on the horizon and concern mounting over deepfaked candidates and fabricated audio clips, the regulator’s consultation document sets out an ambitious framework — while quietly conceding that the technology to enforce it may not yet be up to the job.
The draft, circulated to industry stakeholders this week, would apply to large user-to-user services and search platforms during designated election periods. It proposes that any content depicting real political figures, candidates or party messaging, where that content is wholly or substantially generated by AI, must carry a clear, machine-readable and human-visible label. The obligation would sit alongside existing duties under the Online Safety Act, though Ofcom stresses the labelling regime is a distinct workstream still open to revision.
What the draft actually requires
At its core, the proposal asks platforms to do two things: detect synthetic political content at scale, and attach durable provenance information to it. Ofcom points to emerging standards such as the C2PA content credentials framework as a likely technical backbone, but stops short of mandating any single approach.
- Visible labels on AI-generated or AI-altered political media during election windows
- Embedded, tamper-evident metadata signalling the content’s origin
- Clear appeals routes for users who believe genuine content has been wrongly flagged
- Transparency reporting on detection accuracy and false-positive rates
Crucially, the draft places the detection burden on platforms rather than on the regulator itself. That distinction has already drawn scrutiny. Ofcom would set expectations and audit compliance, but the front-line work of identifying synthetic media would fall to the companies hosting it.
The enforcement problem nobody has solved
The central weakness, critics argue, is that reliable detection of AI-generated content at internet scale remains an unsolved problem. Watermarking schemes can be stripped, and provenance metadata is frequently destroyed the moment a piece of media is screenshotted, re-encoded or re-uploaded across services.
“The honest position is that we cannot currently watermark our way out of this,” said Dr Priya Venkataraman, a synthetic media researcher at the Institute for Digital Provenance. “A C2PA credential survives beautifully until the moment someone takes a photo of their screen and posts it. At that point you have a clean image with no signal in it, and detection models fall back to guesswork.”
That guesswork carries political risk. False positives — labelling authentic footage of a candidate as synthetic — could be weaponised, allowing genuine material to be dismissed as fake. This is the so-called “liar’s dividend”, where the mere existence of deepfakes lets bad actors disown real evidence.
Marcus Eldridge, a regulatory analyst at Whitfield Strategy, warned that the scheme’s credibility hinges on accuracy nobody can yet guarantee. “Ofcom is asking platforms to make high-stakes calls about political speech in real time, during an election, with tools that have meaningful error rates,” he said. “Get it wrong in either direction and you’ve either let a deepfake spread or you’ve suppressed a legitimate campaign message. There is no comfortable margin for error here.”
How it compares internationally
The UK is not acting in a vacuum. The EU’s AI Act includes transparency obligations for synthetic content, and several US states have passed election-specific deepfake laws. But Ofcom’s draft is notable for binding the requirement to platforms via a media regulator with audit powers, rather than relying solely on disclosure duties placed on content creators.
Industry response has been cautious. A representative from one major platform, speaking on background, described the timeline as “extremely tight” given the technical lift involved, and questioned whether smaller services could realistically comply during a compressed election window. Civil liberties groups, meanwhile, have raised concerns about the precedent of regulators shaping which political content gets flagged, however well-intentioned the aim.
Open questions before the rules land
The consultation explicitly invites views on the thorniest issues, including how to handle satire and parody, where the threshold for “substantially AI-generated” should sit, and how labels should persist across re-uploads. Ofcom also seeks evidence on whether mandatory provenance could push the most malicious actors onto unregulated channels entirely.
The regulator has said it will not finalise the framework until after the consultation closes, leaving open the possibility of significant changes — or a phased introduction that begins with transparency reporting before full labelling obligations take effect.
What this means
If adopted, the rules would mark a genuine first: a binding, regulator-enforced labelling regime for synthetic political media in the UK. But the draft reads less like a finished solution and more like an admission of the gap between policy ambition and technical reality. Until detection and watermarking can survive the everyday churn of screenshots and re-uploads, the burden — and the blame — will rest on platforms making imperfect judgements under election-period pressure. For voters, the practical takeaway is sobering: a label, when it appears, will be a useful signal, but its absence will guarantee nothing.
Photo by Markus Winkler on Pexels
