On May 6, 2026, the Paris Prosecutor's Office filed an introductory requisition seeking to place X Corp, X.AI Holdings, xAI, Elon Musk and former CEO Linda Yaccarino under formal investigation on nine criminal charges. The charges range from complicity in distributing child sexual abuse material and non-consensual deepfakes to fraudulent data extraction and — the count that anchors the free-speech debate — facilitating the denial of crimes against humanity through the Grok AI chatbot. (TechPolicy.Press)
The case did not begin with the Holocaust. It began in January 2025 with two complaints: one from National Assembly member Éric Bothorel over opaque algorithm changes, and one from a cybersecurity official alleging that X's recommender system amplified "hateful, racist, anti-LGBTQ" content and distorted democratic debate in France. Prosecutors opened a formal investigation in July 2025, raided X's Paris offices on February 3, 2026, and summoned Musk and Yaccarino in April — neither appeared. (Euronews)
The case for the prosecution is real
It would be dishonest to treat this as a frivolous overreach. In November 2025, Grok generated French-language posts claiming the gas chambers at Auschwitz-Birkenau were built for "disinfection with Zyklon B against typhus" rather than mass murder — textbook Holocaust denial that was viewed roughly a million times before correction. (PBS NewsHour) France's Gayssot Act of July 13, 1990 — article 24 bis of the 1881 press-freedom law — makes contesting the crimes against humanity defined in the 1945 London Charter an offence punishable by up to a year in prison and a €45,000 fine. (The Local) That law has stood for 35 years and survived constitutional review. The deepfake and CSAM counts are graver still. A platform that industrialises the production of such material, then drags its feet on cooperation, invites exactly the scrutiny it is getting. The strongest version of France's position is simple: the harms are concrete, the statutes predate Musk, and no company is above them.
But Europe already built the proportionate tool — and used it
Here is the problem. The European Union spent years constructing a bespoke instrument for precisely this class of harm — systemic risk from very large platforms — and it works through the platform's design, not its executives' liberty. The Digital Services Act subjects X to audited risk assessments, transparency duties and recommender-system obligations, with fines up to 6% of global turnover. In July 2024 the Commission issued preliminary findings that X breached Articles 25, 39 and 40 on deceptive "blue checkmark" design, ad-repository transparency and researcher data access. (European Commission) In December 2025 it converted that into the DSA's first non-compliance fine — €120 million — with deadlines for X to fix the architecture. (European Commission)
That is what proportionate enforcement of algorithmic amplification looks like: regulator-led, focused on systems and transparency, escalating through fines and corrective orders before it ever reaches a courtroom. It targets the design choices that turn individual posts into mass exposure — the actual mechanism the original French complaints described — without asking a criminal court to assign personal intent to a probabilistic text generator.
Where the criminal route strains
The French prosecution, by contrast, bundles genuinely distinct harms — CSAM, deepfakes, data offences, Holocaust denial — into a single nine-count package and attaches them to named individuals. Two problems follow. First, attribution: "facilitating" denial of crimes against humanity through an AI that hallucinated a false claim, then corrected it, is a far less stable legal theory than charging a person who chose to publish that claim. Criminal liability traditionally requires intent or knowledge; a model's output is neither cleanly. Stretching speech statutes written for human authors onto stochastic systems risks bad precedent that outlasts this defendant.
Second, forum: when a 27-state regulator with dedicated platform-governance powers is already mid-enforcement against the same company for the same amplification dynamics, a parallel national criminal probe that could place executives under formal investigation raises the spectre of the very politicisation Musk alleges — even if his bad faith makes him an unsympathetic messenger. The risk is not that X escapes accountability; it is that the manner of accountability hardens into a template where any government displeased with a recommender algorithm reaches for the prosecutor rather than the regulator.
The proportionate path
None of this excuses X. If Grok mass-produced illegal sexual imagery, the deepfake and CSAM charges should proceed on their own merits — those are crimes regardless of platform design. But the algorithmic amplification of hateful content — the spine of the original 2025 complaints — is exactly what the DSA was built to police, and it is policing it. France's interest is best served by letting the systemic-risk machinery run, reserving the criminal law for conduct that is unambiguously criminal and unambiguously intentional. Conflating the two tells every future government that the shortcut to disciplining a platform's algorithm is an indictment. That is a worse internet than the one the Gayssot Act was written to protect.