KOHEN AVOCATS FIRM · PARIS 16
Russian Election Interference in France: What the Paris Deepfake Probe Can Actually Prove

Hassan KOHEN acts in urgent matters, from the police station to the assize court. Initial strategic analysis: €80 incl. VAT, personal reply within 24 hours.

100% confidential · Legal professional privilege · No obligation

Article generated by artificial intelligence, through a process designed and supervised by the firm

Source: French Court of Cassation – Judilibre and Légifrance open data

Paris Bar Police custody, investigation, assize court CNB listing avocat.fr
Hassan Kohen, criminal defence lawyer in Paris
Hassan KOHEN
Attorney at the Paris Bar

Russian Election Interference in France: What the Paris Deepfake Probe Can Actually Prove

Speak to a lawyer 06 89 11 34 45

French prosecutors have opened an investigation into suspected Russian disinformation aimed at possible candidates for the 2027 presidential election. The reported operation used deepfakes, fabricated video reports and false front pages designed to look like established French media. The Paris prosecutor’s office is examining material targeting Édouard Philippe and Gabriel Attal, after an earlier investigation concerning Raphaël Glucksmann. AFP reported the opening of the probe on 18 August 2026, while VIGINUM has publicly described the alleged operation aimed at Gabriel Attal as attributable with a high degree of confidence to the pro-Russian Matryoshka network.

The legal question is narrower than the political accusation. A public authority may identify a foreign influence operation without yet proving who commissioned a particular video, who uploaded it, which account paid to amplify it, or whether the content actually altered a voter’s choice. French law separates those questions. A deceptive representation of a candidate may trigger the specific offence covering algorithmically generated content. A false news publication may engage the Press Act. A coordinated foreign operation may raise national-security issues. An electoral emergency application has its own short statutory window and its own evidential test.

This analysis explains what the Paris investigation can establish, what it cannot assume, and how a candidate or campaign team should preserve evidence before a viral post disappears. The French and European tools are real, but they operate on different clocks. The first task is therefore to translate the geopolitical story into provable acts, identifiable intermediaries and a remedy that matches the moment of the campaign.

I. What the Paris investigation can legally establish before the 2027 vote

A. Why a deepfake is first a concrete offence, not a geopolitical label

The starting point is the content itself. Calling a video “Russian interference” does not prove that it is fabricated, that it depicts a real person without consent, that it was distributed without the required disclosure, or that its circulation was intended to affect the election. Each proposition needs evidence. The fastest legal qualification may nevertheless be domestic and specific: the French offence protecting a person’s representation now expressly includes an algorithmically generated visual or audio content.

Article 226-8 of the French Penal Code, as amended by the SREN law, provides in its first paragraph: Est puni d’un an d’emprisonnement et de 15 000 euros d’amende le fait de porter à la connaissance du public ou d’un tiers, par quelque voie que ce soit, le montage réalisé avec les paroles ou l’image d’une personne sans son consentement, s’il n’apparaît pas à l’évidence qu’il s’agit d’un montage ou s’il n’en est pas expressément fait mention. The next paragraph extends the same treatment to an algorithmically generated visual or sound content representing a person’s image or words where the content is not obviously artificial and is not expressly identified as such. If the offence is committed through an online public communication service, the statutory maximum increases to two years’ imprisonment and a €45,000 fine.

That text creates several questions that a prosecutor must answer rather than leave to political rhetoric. Is the person recognisable? Was the image or voice used without consent? Did the publisher make the artificial nature obvious or expressly disclose it? Was the content made available to the public or to a third party? Who performed the act of publication? An account that merely comments on a genuine recording may raise different issues from an account that supplies a synthetic voice, a fake interview, a fabricated newspaper cover and an impersonating domain name.

The rule does not require the prosecutor to prove that the content changed the election result. It protects the person’s representation. That makes it valuable at the beginning of an investigation, when the political impact is difficult to quantify. It also means that a candidate should not wait for a poll to show damage. The evidential file should focus first on the original file, its publication, its lack of disclosure and the chain of accounts that distributed it.

The leading pre-amendment authority remains Cour de cassation, Criminal Chamber, 30 March 2016, no. 15-82.039. In a case concerning a television report, the decision records the principle that le délit d’atteinte à la représentation de la personne ne réprime pas le montage en tant que tel, mais en ce qu’il tend à déformer de manière délibérée des images ou des paroles. The Court upheld the finding that a visibly edited report, which did not manipulate the meaning of the recorded words and images, did not satisfy the offence. The case is not a licence to circulate a deceptive AI video. It is a warning that editing alone is not the legal harm: the prosecution must identify the distortion and the way the public was led to mistake the representation for an authentic one.

That distinction matters in the French campaign context. A fake video may contain no technically perfect face swap. It may use an authentic speech, remove a sentence, add a false voice-over and present the result as a new statement. It may place a candidate’s genuine photograph beside a fabricated headline. It may imitate the visual identity of a broadcaster rather than the candidate’s face. Article 226-8 is strongest where the person’s words or image are themselves falsely represented, but the same factual package may also support other qualifications: identity usurpation, defamation, threats, harassment, fraud, document forgery or an electoral offence.

Article 27 of the Press Act supplies a different test. Article 27 of the Law of 29 July 1881 states: La publication, la diffusion ou la reproduction, par quelque moyen que ce soit, de nouvelles fausses, de pièces fabriquées, falsifiées ou mensongèrement attribuées à des tiers lorsque, faite de mauvaise foi, elle aura troublé la paix publique, ou aura été susceptible de la troubler, sera punie d’une amende de 45 000 euros. The offence is not a general prohibition on inaccurate political opinion. It requires a “new” or a fabricated document, bad faith and actual or potential disturbance of public peace. The public prosecutor must also respect the special procedural rules of press offences.

The boundary was drawn by Cour de cassation, Criminal Chamber, 13 April 1999, no. 98-83.798. The published summary states: Ne constituent pas une nouvelle, au sens de l’article 27 de la loi du 29 juillet 1881, les affirmations et reviews, tendancieux ou mensongers, portant sur un fait déjà révélé. In practical terms, a post that comments aggressively on an already known event may be defamatory or otherwise unlawful without automatically being a new false report under Article 27. A fabricated “breaking” story about a candidate’s health, a fake broadcaster page announcing a corruption finding, or a synthetic interview that never occurred is easier to frame as a new assertion of fact.

The electoral offence is narrower again. Article L. 97 of the Electoral Code provides: Ceux qui, à l’aide de fausses nouvelles, bruits calomnieux ou autres manoeuvres frauduleuses, auront surpris ou détourné des suffrages, déterminé un ou plusieurs électeurs à s’abstenir de voter, seront punis d’un emprisonnement d’un an et d’une amende de 15 000 euros. The prosecution must connect the false news or manoeuvre to votes or abstention. A campaign team should therefore record not only that a post was false, but also its timing, target audience, geographic reach, paid distribution, calls to vote or abstain, and relationship to the candidate’s electoral support.

These provisions can overlap, but they do not merge. Article 226-8 asks whether a person’s representation was deceptively published without consent. Article 27 asks whether a bad-faith false news publication was capable of disturbing public peace. Article L. 97 asks whether false news or a fraudulent manoeuvre surprised or diverted votes or caused abstention. The same deepfake can satisfy one test, several tests, or none of them if the available proof does not establish the required element. A responsible complaint should present alternative legal qualifications and identify the evidence for each one.

The distinction also protects political speech. A campaign may lawfully criticise a candidate, repeat a disputed allegation in a clearly attributed way, parody a public figure, or publish an edited extract that is transparently presented as an edit. The legal risk begins when presentation makes the public believe that a fabricated image, voice or media report is authentic. The investigation must examine context, disclosure, intent, distribution mechanics and the ordinary viewer’s capacity to understand what was being shown.

B. When Russian attribution becomes an offence against the Nation

Attribution is a separate evidential project. VIGINUM’s public methodology is useful because it describes a foreign digital interference operation through four cumulative features: manifestly inaccurate or misleading allegations, artificial or automated and massive deliberate dissemination, a risk to the fundamental interests of the Nation, and the direct or indirect involvement of a foreign actor. VIGINUM explains that it works through open-source investigation and analysis of information operations. It also states that it does not use intelligence techniques and does not perform fact-checking in the ordinary sense. Its attribution is an operational assessment, not a criminal conviction.

That difference should be kept visible in every public statement. A VIGINUM assessment can help the Paris prosecutor locate the relevant accounts, identify a mode of operation and understand how domains, social profiles and distribution channels fit together. It does not by itself prove that a particular person ordered the content, that the Russian state controlled every intermediary, or that a criminal offence has been committed by every account that shared the material. Investigators must convert the operational pattern into admissible proof: account creation data, payment records, instructions, domain registration, device or server links, coordinated timestamps, repeated assets, messages between operators and the content of the instructions.

Article 411-5 of the Penal Code may become relevant only where that evidence shows more than foreign origin. The article provides: Le fait d’entretenir des intelligences avec une puissance étrangère, avec une entreprise ou organisation étrangère ou sous contrôle étranger ou avec leurs agents, lorsqu’il est de nature à porter atteinte aux intérêts fondamentaux de la nation, est puni de dix ans d’emprisonnement et de 150 000 euros d’amende. The text requires a relationship or intelligence with a foreign power, organisation or agent, and conduct capable of harming the Nation’s fundamental interests. The words do not turn every foreign narrative, foreign-funded media project or foreign-linked account into espionage.

For a presidential campaign, the relevant interests may include the integrity of republican institutions and the ability of citizens to make an informed electoral choice. But the prosecutor still has to prove the connection. It is not enough to show that a false claim favoured a geopolitical position associated with Moscow. The file should show who had the foreign link, how the link operated, what instructions or resources were supplied, and why the operation was capable of harming a protected national interest. The threshold is serious precisely because the offence carries a ten-year maximum sentence.

Organised-network theories must be handled with the same care. Article 450-1 of the Penal Code now defines an association of wrongdoers as tout groupement formé ou entente établie en vue de la préparation, caractérisée par un ou plusieurs faits matériels, d’un ou plusieurs crimes ou d’un ou plusieurs délits punis d’au moins cinq ans d’emprisonnement. The same article sets a five-year punishment for participation where the planned offences are misdemeanours carrying at least five years. A network of accounts does not automatically meet that threshold merely because it is coordinated. The ordinary online deepfake offence under Article 226-8 carries a lower maximum, so prosecutors would need to identify another sufficiently serious planned offence before relying on this association provision.

That legal arithmetic is important in the present story. A coordinated operation may be operationally real and politically harmful without satisfying the elements of Article 411-5 or Article 450-1. Conversely, a campaign may begin as a representation offence and reveal, through payment flows or foreign instructions, a wider criminal plan. The correct approach is progressive: secure the material, identify the domestic offences that are already evidenced, then test whether the foreign relationship and the network’s purpose justify a national-security or organised-crime qualification.

Article 27 has its own procedural trap. In Cour de cassation, Criminal Chamber, 30 January 2019, no. 17-87.127, concerning alleged false news connected with a disputed political-financing document, the decision states that la personne qui se prétend lésée par le délit de publication de fausses nouvelles prévu par l’article 27 alinéa 1 de la loi précitée ne peut mettre en mouvement l’action publique soit en se constituant partie civile soit en usant de la procédure de citation directe et que ce droit est réservé au seul ministère public. A candidate can report the facts, preserve evidence and seek advice on civil or urgent remedies, but a complaint should not promise that a private party can personally launch an Article 27 prosecution through any procedural route.

The foreign-investigation file should therefore separate three layers. The first layer is the media object: the video, audio, image, headline or domain. The second is the distribution system: accounts, reposts, bots, paid advertisements, language variants, targeting and amplification. The third is attribution: the infrastructure, financial, organisational and communicative links to a foreign actor. Each layer may be proven by different witnesses and different technical evidence. Keeping them separate makes it possible to act quickly on the first layer without overstating what has been established about the third.

That discipline is also essential for international media. The safest formulation is that French prosecutors are investigating suspected Russian interference and that VIGINUM has attributed an operation with a stated degree of confidence. The unsafe formulation is that Russia has been judicially found responsible for every post or that the operation has already altered the election. The first is a sourced report about an investigation; the second would be an unsupported conclusion. French procedure can move rapidly, but it still demands a link between facts and legal elements.

II. What victims, platforms and investigators can do before the legal clock closes

A. The 48-hour electoral takedown route and the evidence file

The electoral emergency route is designed for a particular moment. Article L. 163-2 of the Electoral Code applies during the three months before the first day of the month of general elections and until the relevant round is acquired. It targets inaccurate or misleading allegations about a fact that may alter the sincerity of the coming vote when they are disseminated deliberately, artificially or automatically and massively through an online public communication service. The judge may prescribe proportionate and necessary measures to stop the dissemination at the request of the public prosecutor, a candidate, a political party or group, or any person with an interest in acting. The judge must rule within forty-eight hours, and the appeal court has the same period.

The operative wording is worth reading carefully: lorsque des allégations ou imputations inexactes ou trompeuses d’un fait de nature à altérer la sincérité du scrutin à venir sont diffusées de manière délibérée, artificielle ou automatisée et massive. This is not a general emergency button for every unpleasant post. The applicant must show a factual assertion, its inaccurate or misleading character, a distribution pattern that is deliberate, artificial or automated and massive, and a real capacity to affect the sincerity of the upcoming vote. The court can order measures against the relevant online service or intermediary. The procedure is urgent, but the evidential burden remains real.

For the candidate, the first six hours should be devoted to preservation rather than public repetition. Save the original URL, the account identifier, the publication time displayed in the relevant time zone, the full text, the audio or video file, the accompanying image, the number of views and shares, the comments and any paid-promotion disclosure. Take screenshots that include the account name and URL, but also preserve the underlying file and record its cryptographic hash. A screen recording can show how the post was reached and whether it redirects through a cloned media page. An independent bailiff or digital-forensics professional can add date and integrity evidence when the expected litigation value justifies the cost.

The file should also record the elements that turn virality into an electoral argument. Identify the candidate named in the content, the alleged event, the election and the date of the vote. Map the first account, the first major reposts, any network of newly created accounts, paid advertisements, language versions and targeting indicators. Preserve evidence that the content was presented as a genuine report: an imitation logo, a cloned headline, a fabricated journalist account, a synthetic voice or an absence of an AI label. If the message tells voters not to vote, to vote for a named candidate or to believe that a candidate has withdrawn, record those instructions precisely.

A careful applicant should put the material into a chronology. Column one identifies the post and its URL. Column two identifies the person or organisation represented. Column three states the false factual assertion and the source disproving it. Column four records the distribution method, reach and timing. Column five links the act to the requested remedy. This format allows a lawyer to select the strongest examples instead of presenting a court with hundreds of unindexed screenshots. It also protects the campaign from accidentally amplifying the most harmful content by reproducing it in a public statement.

The statutory clock makes preparation before the election window decisive. If the content appears outside the period covered by Article L. 163-2, an applicant cannot simply insert it into that procedure by calling it electoral. Other remedies may remain available: a report to the public prosecutor, a complaint concerning Article 226-8, a defamation strategy, a platform notice, a data-protection complaint where personal data are misused, or an ordinary interim application when its conditions are met. Each route has its own standing, limitation and proof requirements.

The criminal investigation can obtain information that a campaign team cannot obtain directly. Article 60-1 of the Code of Criminal Procedure states that the prosecutor or judicial police may require relevant information from a person, private or public organisation or public administration, including information from an information system or nominative data processing. Its opening formula is: Le procureur de la République ou l’officier de police judiciaire … peut, par tout moyen, requérir de toute personne, de tout établissement ou organisme privé ou public ou de toute administration publique qui sont susceptibles de détenir des informations intéressant l’enquête. During a preliminary investigation, Article 77-1-1 provides a comparable power, including the delivery of digital information.

These provisions help investigators request account records, hosting information, domain data, advertising information and other material held by intermediaries. They do not guarantee immediate disclosure by a platform established abroad. The request may have to follow a specific international cooperation channel, satisfy a foreign legal standard or identify a sufficiently precise account and time period. That is why a campaign should preserve every public trace while it is still available. The prosecutor can seek provider-held evidence later, but a deleted post, an expired story or an account that changes its handle can make the request slower and less certain.

The complaint should state what is known and what remains to be established. It can say that a video appears algorithmically generated, that it was published without a disclosure, that it imitates a broadcaster and that it was massively reposted during a campaign. It should not state as fact that a named foreign service ordered the post unless the evidence supports that assertion. A request for preservation and identification is stronger when it is precise, neutral and tied to the suspected offences.

Finally, a campaign should distinguish removal from correction. A platform takedown may stop further distribution but may not identify the operator, preserve the evidence or repair the candidate’s reputation. A public correction may alert voters but also increase the reach of the false story. The legal strategy should therefore coordinate the platform notice, the judicial application, the criminal report and the communications decision. The quickest public response is not always the response that gives investigators the best proof.

B. Why France must combine domestic remedies with EU platform and cross-border cooperation

The French court is only one part of the response. The account may be hosted in one country, the domain registered in another, the payment processor established in a third, and the audience spread across France and the European Union. A candidate seeking to stop the content must address the platform that can actually remove or demote it, while a prosecutor seeking attribution must address the provider that holds the logs. A national complaint and a platform notice should be coordinated, not treated as substitutes.

Under the Digital Services Act, very large online platforms and search engines must assess and mitigate systemic risks, including risks to civic discourse and electoral processes. The Commission’s framework includes crisis-response tools, while national coordination in France involves Arcom. Arcom’s published electoral guidance asks platforms to explain what users can report, adapt their services to the election context, deploy adequate human and technical resources, increase moderation transparency, highlight reliable official sources and remove advertisements that amount to unlawful electoral propaganda when reported. Those recommendations do not decide criminal liability, but they provide a practical escalation route when an operation is spreading through a platform’s recommender or advertising systems.

VIGINUM’s own public description confirms the value of this bridge. It supplies information useful to Arcom and electoral authorities, studies information operations through open sources and shares technical elements concerning inauthentic activity with major platforms. The agency’s account of the threat also identifies generative AI, media and administration typosquatting, proxy networks and commercial intermediaries as recurring features. A campaign team should therefore send the platform more than a statement that “this is fake.” It should identify the impersonated media outlet, the repeated assets, the linked accounts, the artificial amplification, the relevant language variants and the exact policy or legal harm.

The AI Act adds a transparency layer. Article 50 of Regulation (EU) 2024/1689 requires deployers of an AI system that generates or manipulates image, audio or video content constituting a deepfake to disclose that it was artificially generated or manipulated, subject to the regulation’s exceptions and allocation of responsibility. A label is not a defence to a deceptive campaign if other offences are present, but the absence of a clear disclosure can be relevant to Article 226-8 and to the platform’s assessment of misleading conduct. Conversely, the existence of AI-generated content alone does not prove criminal intent. The content, the context and the distribution remain decisive.

International cooperation is often the difference between an attribution hypothesis and a charge. Investigators may need subscriber information, registration records, advertising invoices, payment trails, cloud logs, IP connection data or communications between operators. For European providers, the available judicial cooperation framework may allow a direct production or preservation route depending on the data and the date of application. For providers outside the European Union, mutual legal assistance, a preservation request or a provider’s emergency process may be necessary. The request should identify the account, URL, time range, suspected offence and urgency. A vague request for “all information about Russian disinformation” is unlikely to produce useful evidence.

That cross-border dimension also affects publication risk. A French candidate may be targeted by a post published abroad, but French courts may still have jurisdiction over harm occurring in France or conduct aimed at French voters, subject to the applicable jurisdictional and press-law rules. A foreign actor may invoke freedom of expression, satire or political criticism. The response must therefore explain why the specific content is a deceptive factual representation, why the French electoral process is implicated, and why the requested measure is proportionate. Removing every controversial opinion would be neither legally sound nor compatible with the protection of political debate.

The European human-rights framework reinforces that proportionality exercise. Political speech receives strong protection, including speech that shocks or disturbs. A court is more likely to grant a targeted order where the applicant identifies an impersonated candidate, a false fact, a synthetic representation, a coordinated artificial distribution pattern and a time-sensitive electoral risk. It is less likely to approve a broad order against an undefined “Russian narrative” or an entire class of criticism. The remedy should name the content and the necessary action: remove a specific URL, disable a cloned account, preserve data, disclose the artificial nature of a file or stop a paid campaign.

The case law on manipulated representations supports this measured approach. The decision in no. 15-82.039 did not treat every edit as a criminal montage; it examined whether the edit distorted the meaning of the recorded words and images and whether the audience could see that it was a montage. The current Article 226-8 expressly covers algorithmic content, but the underlying need to identify the deceptive representation remains. A campaign should obtain a technical report where the alleged deepfake is disputed, while also showing the surrounding presentation that made the content appear authentic.

There is a second proportionality issue in public communications. A candidate may have a strong interest in warning voters, but repeating a false claim in full can give it a second life. The better practice is to identify the false content by a stable reference, publish a short correction that states the verified fact, direct supporters to the campaign’s authentic channels, and provide the preserved material to counsel and investigators. If the false item impersonates a newspaper or broadcaster, the genuine outlet’s confirmation can be valuable evidence, but it should be linked without reproducing unnecessary screenshots of the fabricated headline.

For an international legal team, the operational sequence is straightforward. First, preserve the source and calculate an integrity hash. Second, capture the account, URL, timestamp, reach and amplification path. Third, obtain independent confirmation of the false fact or the synthetic nature of the voice or image. Fourth, notify the platform and request preservation under its emergency or legal process. Fifth, file a targeted report with the French prosecutor and identify the relevant offences without collapsing them into a geopolitical conclusion. Sixth, assess whether the Article L. 163-2 window is open and whether the deliberate, artificial and massive dissemination test can be met. Seventh, coordinate any EU or foreign request for provider data. Eighth, decide publicly whether correction or silence better protects the electoral process.

That sequence keeps the domestic and international tracks aligned. The French court can address the urgent harm. The prosecutor can seek data and test the criminal qualifications. Arcom and the platform can address the service-level risk. VIGINUM can document the foreign operating pattern. Foreign authorities or providers can supply the evidence that connects the infrastructure to the operators. No single institution has to prove every element alone, but each must be given a precise question and a reliable record.

Conclusion

The Paris probe is significant because it moves the story from a media warning to a legal investigation. Its outcome will depend on whether the alleged deepfakes can be tied to a deceptive representation, a bad-faith false publication, an electoral manoeuvre, a foreign relationship or a coordinated plan through evidence that meets each relevant test. VIGINUM attribution, platform action and political reporting are important, but they are not interchangeable with proof in court. Candidates and campaigns that preserve the original files, map the amplification and choose the right remedy early will have the best chance of protecting both the vote and the integrity of the investigation.

📄 Official circular