On May 2, 2026, President Vladimir Putin signed Federal Law No. 130-FZ, amending Russia's electoral legislation to govern when a person's image or voice — including versions generated with "information technologies," the statute's term for AI — may appear in election campaign materials. The Federation Council approved the bill on April 29, 2026, after the State Duma passed it in second and third readings on April 21. As a baseline, the law bars campaigns from deploying a person's likeness or synthesized voice, with three carve-outs: an electoral association may use the image and voice of the candidate it nominated; a candidate may use their own; and any other depiction requires the prior written consent of the depicted person, who must be a Russian citizen aged 18 or older. Materials that feature a person affiliated with a "foreign agent" must carry a disclosure to that effect.
The case for consent and disclosure
The strongest version of the regulator's argument is genuinely strong. Generative tools now make it trivial to fabricate a candidate's endorsement, put invented words in a rival's mouth, or manufacture a crowd of synthetic supporters — and voters cannot reliably tell the difference at a glance. A consent requirement protects a real, longstanding interest: control over one's own face and voice. A disclosure requirement, in principle, gives audiences the context to discount manipulated content rather than banning it outright. Importantly, Russia did not enact a blanket prohibition on AI in politics; it conditioned use on permission and labeling. Framed narrowly — don't impersonate people without their say-so, and tell readers when content is synthetic — this is closer to proportionate, transparency-first regulation than to censorship. That is the model much of the democratic world is converging on.
A reading that swallows the rule
The problem is what surrounds the text. Two months before Putin signed the law, on February 4, 2026, Russia's Supreme Court upheld a regional election commission's ban on a Communist Party (KPRF) leaflet that used AI-generated, non-specific human figures under the slogan "We value everyone," according to Vedomosti's reporting on the ruling. The court read the concept of "physical person" broadly enough to cover AI-generated typified human images — generic, invented figures depicting no real individual — treating them as representations of actual people. The KPRF called the decision an "undeclared administrative barrier" and said it would appeal to the Constitutional Court.
That reading matters because it pre-loads the new statute with maximal scope. A consent rule aimed at non-consensual impersonation is defensible. A consent rule that, via judicial interpretation, also captures illustrative, fictional, or composite imagery — content that impersonates no one — is something else: a constraint on ordinary political design work. If a stock-style AI illustration of "a voter" can be banned because a court deems it a depiction of a person, the rule stops being about deception and starts being about who controls the visual vocabulary of a campaign.
The foreign-agent tell
The disclosure provision is where proportionality breaks down most clearly. On its face, labeling is the least restrictive tool — more speech, not less. But Russia's foreign-agent framework is not a neutral transparency regime. Under the law expanded in December 2022, the "foreign agent" tag is applied to independent journalists, opposition figures, and organizations deemed under "foreign influence," and it already carries onerous labeling and reporting burdens. Bolting that designation onto campaign-material rules means a candidate who features, quotes, or even visually references a tagged person must broadcast the stigmatizing label — a built-in deterrent against associating with disfavored voices. The consent-and-disclosure architecture that reads as light-touch in the abstract becomes, in this context, another node in an apparatus for marginalizing dissent.
What proportionate looks like
The contrast with peer regimes is instructive. The EU's AI Act requires deployers to disclose AI-generated or manipulated content, and roughly 30 U.S. states have enacted laws targeting deceptive synthetic media in political communications — overwhelmingly through disclosure mandates and narrow deception standards rather than image bans. Those approaches share a discipline Russia's framework lacks: they target the harm (deceiving voters about authenticity or authorship) rather than the medium (the use of AI imagery as such), and they operate inside systems where courts and election bodies are not instruments of incumbent advantage.
From a pro-innovation standpoint, the lesson is not that consent and disclosure are wrong — they are sensible defaults that good regulation should adopt. It is that the same words do very different work depending on the enforcement environment. A consent-plus-labeling rule administered by independent referees protects candidates and voters alike. The identical rule, administered by a court that treats invented figures as real persons and a state that wields "foreign agent" as a scarlet letter, becomes a discretionary lever — one that can be relaxed for the incumbent association using its own candidate's synthetic likeness and tightened against everyone else.
Russia has, in effect, written a reasonable-sounding deepfake statute and embedded it in machinery that converts reasonableness into control. The cautionary takeaway for democracies drafting their own rules is precise: keep synthetic-media regulation tied to demonstrable deception, define "depiction" narrowly enough to spare fictional and illustrative content, and never let a transparency mandate double as a loyalty test.