What the committee approved
On 29 September 2026 the Chamber of Deputies' Comisión de Economía, Comercio y Competitividad approved, by 18 votes to 6, a dictamen from President Claudia Sheinbaum's initiative amending the Ley Federal del Derecho de Autor. According to Ovaciones, it amends article 114 Octies and adds articles 114 Nonies and 232 Septies. It creates "responsabilidad secundaria" for internet service providers and social networks. The Chamber's own Gaceta lists the dictamen under the title "en materia de responsabilidad secundaria para proveedores de servicios de internet". Press coverage links it to the removal of works, audio or voices altered with AI in violation of copyright. Morena and its allies voted in favour. PAN, PRI and MC voted against, and two PAN reservations were rejected.
Mexico still has no general AI law. ITSitio reported in August 2026 that general-law initiatives sit in committee and that rules are advancing sector by sector. This dictamen is one of those sector measures. It regulates AI through copyright enforcement against intermediaries, not through any rule about AI systems themselves.
The strongest case for the reform
The harm is real. A cloned voice or a synthetic performance can be distributed to millions of people before its victim finds out. Individual performers cannot chase thousands of uploads across platforms. Secondary liability is also an established legal tool. Per UnoTV, it applies only where a provider knows of an infringement and also contributes to it, induces it, can control it, or profits directly from it. That is narrower than strict liability. The reform builds on the April 2026 Chamber vote on voice and image. Excélsior reported that vote at 368 in favour on the particular vote. That text requires a written agreement for any AI-based transformation or impersonation of a person's voice or image, and it carves out parody, satire and creative imitation that does not impersonate. The government also presents the new rules as harmonisation with the T-MEC, according to El Universal reporting carried by Diario de Yucatán. Rightsholders asking for a duty to act on infringing AI imitations are making a reasonable demand.
Where the design falls short
The problem is procedure, not purpose. Reporting on the dictamen describes a duty to remove infringing content "de manera expedita", with no deadline in hours or days. None of the coverage I read mentions a court-order requirement, a counter-notice mechanism or a safe harbour for good-faith removal. Meanwhile the penalties are heavy. IMPI could fine non-compliant providers up to 40,000 UMA, which the El Universal report puts at over 4.3 million pesos. UnoTV reports a range from 117,310 pesos to more than 4.6 million pesos. The two outlets give different ceilings, and the dictamen text should settle which is right.
Set a vague duty beside a large fine and a platform's cheapest option is to delete first and ask questions never. PAN deputy Marcelo Torres Cofiño made this point in committee, according to TechnoNoticias, saying platforms will block content quickly to avoid sanctions. A parody that closely imitates a politician's voice is exactly the kind of content that gets swept up. Judging whether it counts as impersonation takes context that an automated filter lacks.
The repeat-infringer provision raises a second concern. Providers must adopt policies to suspend or permanently end the service or accounts of repeat infringers. UnoTV quotes Torres Cofiño asking who decides recidivism, by what procedure and with what guarantees, warning that someone could lose internet access before a court rules. Legislator María Angélica Granados raised due-process objections to disconnection, as TechnoNoticias reports. Cutting a person's connection is a far larger intervention than removing a file. It affects access to work, education and public services.
The treaty the government cites asks for more
The T-MEC harmonisation argument cuts the other way. Chapter 20 of the USMCA requires enforcement procedures to be "applied in such a manner as to avoid the creation of barriers to legitimate trade and to provide for safeguards against their abuse" (Article 20.78.1). It also says each Party shall take into account "the need for proportionality between the seriousness of the infringement of the intellectual property right and the applicable remedies and penalties, as well as the interests of third parties" (Article 20.78.5). A regime that offers a fast route to removal and a heavy fine but no abuse safeguard is hard to square with that language. Proportionality is a treaty commitment as well as a policy preference.
What a better version looks like
None of this requires abandoning the reform. Four additions would fix most of the problems:
- A defined removal window after a substantiated notice, with a clear notice standard.
- A counter-notice route and reinstatement if the claimant does not pursue the matter.
- Penalties for knowingly false notices, so the system cannot be used for harassment or to suppress criticism.
- A judicial or IMPI order, with a hearing, before any account is permanently terminated for recidivism.
The April reform already shows the model: protect consent and exclude parody that does not impersonate. The new article 114 provisions should carry the same carve-out into the takedown text. The two reservations the committee rejected, which sought to protect public debate, were an easy place to start.
What to watch
The dictamen went to the Chamber's leadership for scheduling on the floor, and the Gaceta of 7 October lists it with reservations. I could not confirm the final floor vote or any Senate action, so readers should check the Gaceta for the latest status. Mexico can still build a framework that tackles AI imitation without making platforms into automatic censors. The amendments that matter are the cheap ones: timelines, counter-notice and judicial review. Leaving them out would turn a defensible anti-impersonation measure into a general incentive to over-remove.