The May 8 Escalation
On May 8, 2026, Paris Prosecutor Laure Beccuau announced that France's preliminary inquiry into X and Elon Musk had been elevated to a full instruction judiciaire — a formal criminal investigation in which an examining magistrate can compel evidence, issue international letters rogatory, and ultimately seek indictments. The probe now spans nine charge categories: complicity in possessing and distributing child sexual abuse material, fraudulent and unlawful data collection, inadequate data security, fraudulent extraction of data from automated systems, violation of correspondence secrecy, falsification of automated data processing operations, distribution of nonconsensual sexual deepfakes, platform administration enabling illicit transactions, and Holocaust denial facilitation via Grok AI.
This did not arrive without warning. Initial complaints were filed on January 12, 2025 by French MPs Éric Bothorel and Arthur Delaporte; Paris formally opened a criminal investigation in July 2025; raids on X's Paris offices followed on February 3, 2026; Musk and former CEO Linda Yaccarino were summoned for voluntary interviews on April 20, 2026 — neither appeared. The May escalation gives prosecutors compelled-process tools they previously lacked.
The Case That Is Hardest to Dismiss
Two charge categories rest on solid ground and deserve to be taken seriously before any critique.
On deepfakes: an analysis of images generated by Grok on January 5–6, 2026 found the tool producing roughly 6,700 sexually suggestive or nudified images per hour — 84 times more output than the top five dedicated deepfake websites combined. A review of 20,000 images from the preceding weeks found 2% appeared to depict minors; a separate forensic analysis of 800 recovered images by Paris-based AI Forensics found approximately 10% contained what appeared to be minors in sexual situations. X restricted Grok's image-editing capability on January 14, 2026 — twelve days after French ministers first flagged the tool to prosecutors. The timeline is not flattering.
On Holocaust denial: French law (Article 24 bis of the Press Law of July 29, 1881) criminalizes contesting crimes against humanity. When Grok generated French-language posts claiming Auschwitz gas chambers were designed "for disinfection with Zyklon B against typhus" rather than mass murder, it produced content that would be illegal for any French publisher. Platform complicity in that output is a legitimate inquiry — and prosecutors have stated that "the functioning of the AI will be examined" as part of the probe.
The Harder Charge: Algorithmic Manipulation
The most legally novel — and most contested — allegation is "falsification of automated data processing system operations" under Article 323-2 of the French Penal Code, which carries up to five years imprisonment and €150,000 in fines. Prosecutors allege X's algorithm was deliberately altered to amplify hateful and far-right political content, spreading material intended to "skew the democratic debate in France."
The strongest version of this argument is not trivial. Algorithmic curation is not neutral; it is an editorial act that operates at scale. If a platform's recommendation engine systematically suppresses certain viewpoints while amplifying others in ways that are opaque, unaudited, and potentially intentional, society has a legitimate interest in scrutiny. France's Digital Services Act implementation gives ARCOM, as the national Digital Services Coordinator, real powers over systemic risks — including algorithmic risk assessments that very large online platforms must publish under DSA Article 34.
But Article 323-2 was written for hacking and sabotage, not for content curation disputes. The statute targets "obstruction or falsification" of a system — acts directed at a system from the outside, or deliberate insider sabotage. Applying it to a platform's own algorithmic choices stretches the provision beyond its natural scope. Criminal liability for editorial algorithms would require showing not just that amplification occurred, but that it constituted a fraudulent falsification — not a misguided, or even manipulative, editorial policy. That evidentiary bar is high. A prosecution that fails on this charge while succeeding on deepfakes could still reshape European law, but it risks muddying the more legitimate parts of the case.
Jurisdictional Friction
There is a structural tension in France's approach that neither side in this debate fully acknowledges. Under the Digital Services Act, the European Commission holds exclusive enforcement authority over Chapter III, Section 5 obligations for Very Large Online Platforms — the systemic-risk provisions most directly applicable to algorithmic manipulation. ARCOM enforces other DSA chapters but cannot unilaterally pursue the systemic-risk provisions against X.
The Commission has run its own parallel DSA investigation into X for nearly two years, and expanded its scope to include algorithmic amplification in January 2026. French prosecutors separately alerted the US Department of Justice in March 2026. These parallel tracks — French criminal law, EU administrative enforcement, and potential US cooperation — could produce inconsistent findings or inadvertently shield X if any one track stumbles. Coordination between Paris and Brussels is not optional; it is essential to whether this becomes a precedent or a procedural tangle.
Proportionality and What Comes Next
The breadth of nine charge categories — sweeping from CSAM to data extraction to correspondence secrecy — risks making the probe appear maximalist. X has characterized it as "law enforcement theater designed to achieve illegitimate political objectives," which is self-interested, but the counter-argument cannot rest solely on the strength of the Grok charges. Investigations that sweep broadly have a weaker claim to legitimacy than those that stay disciplined.
The instruction judiciaire gives French prosecutors tools they now need: compelled testimony, server access requests, and the full weight of international judicial cooperation. Neither Musk nor Yaccarino appeared voluntarily in April; compelled process changes the calculus. Whatever the outcome, France has moved a significant marker: this is the first time any EU member state has formally named algorithmic manipulation as a theory of criminal liability against a major platform. Europe's algorithmic accountability debate will be shaped by whether Paris can make that charge stick.