On August 28, 2026, a bill numbered 4362-D-2026 was introduced in Argentina's Chamber of Deputies. According to the Digital Policy Alert tracker, it creates a regime for the "identification and labelling of digital content generated or modified by artificial intelligence." It applies to public and private entities that produce, distribute or exhibit such content. The tracker lists platform intermediaries hosting user-generated content, e-commerce services and AI developers among the regulated activities. As of the tracker's last update it remains under deliberation. The excerpt available to us does not specify penalties or sponsors, and we could not read the full bill text, so this analysis addresses the design choices the bill will face rather than provisions we cannot confirm.
The strongest case for labelling
The case for the bill is real. Synthetic voices, faces and video can now be produced cheaply, and the harms are concrete: fake endorsements in advertising, impersonation of public figures, and fabricated media during elections. A label is a light-touch remedy. It doesn't ban anything and doesn't require anyone to prove a statement false. It gives the audience one more piece of context. Argentina also has a legitimate interest in acting: its main data protection statute, Ley 25.326, dates from 2000, well before generative AI, and defines personal data broadly as information about individuals, which leaves open how a synthetic likeness should be treated.
Disclosure rules are also the more proportionate end of the AI regulatory spectrum. Labelling is a far smaller intervention than licensing, pre-approval or liability for model outputs. On this point, we think the bill is aimed at the right kind of problem.
Where the design risk sits
The trouble is not the label. It is who must apply it and what they must do to comply. The tracker says the bill reaches intermediaries that host user-generated content. That matters because a platform does not create the content its users upload. If the law is read as requiring platforms to detect AI-generated material and label it themselves, it becomes a monitoring mandate, with three predictable effects.
- Detection is unreliable. Detecting synthetic media at scale produces both misses and false positives. A rule that penalises an unlabelled item invites platforms to over-label or over-remove to be safe.
- Small services bear the cost. A large platform can build classifiers and review teams. A local marketplace or a startup hosting comments cannot, and e-commerce sellers are named in the scope alongside AI developers.
- Legitimate speech gets caught. Satire, art, accessibility tools and light editing (noise reduction, translation, a filter) all "modify" content with AI. A definition broad enough to cover "AI-modified" material could sweep in routine editing.
Argentina already has a doctrine for this
Argentina does not start from a blank slate on intermediary responsibility. In Rodríguez, María Belén v. Google, decided on October 29, 2014, the Supreme Court declined to impose strict liability on search engines. As summarised by Stanford's WILMAP, the Court held that liability generally requires notice from a public authority that content is unlawful, with an exception for gross and manifest harm, and it rejected any obligation to filter in advance so that infringing links would not reappear. That case concerned search engines rather than social platforms or AI tools, so it does not control this bill. It does, however, show how the country's highest court has weighed the free-expression costs of proactive monitoring.
A labelling law that respects that reasoning would place the primary duty where the knowledge is: on the producer or developer that generates the content, and on the user or advertiser who publishes it as their own work. Platforms would then have a narrower, workable role. They would preserve and display labels that arrive with the content, provide a way for uploaders to declare AI use, and act on notice for manifestly deceptive material. It would not require them to guess.
The international benchmark cuts both ways
The European Union is the obvious comparison. According to the European Commission's page on its code of practice for AI-generated content, the AI Act's transparency obligations under Article 50 apply from 2 August 2026. Providers must ensure that outputs are marked in a machine-readable format and detectable as artificially generated. Deployers must disclose deepfakes and AI-generated text on matters of public interest, unless there has been human review and editorial responsibility. Two features are worth borrowing. First, the duty is split between the party that builds the generator and the party that deploys it, which is the more effective division of labour. Second, there is an editorial-responsibility exemption, which protects ordinary publishing.
The lesson is not to copy Brussels. The EU's framework sits alongside a large compliance apparatus that Argentine regulators and small firms may not be able to replicate. The lesson is that a labelling duty works best when it is tied to a specific role, a specific type of content (realistic deepfakes and matters of public interest, not all AI-assisted material) and a specific exemption for human editorial control.
What deputies should settle before a vote
Because the bill is still in committee-stage deliberation, the drafting questions are open, and a few of them will decide whether it helps or harms.
- Penalties. The tracker excerpt does not specify them. Sanctions should attach to knowing failure to label by the producer or publisher, not to a platform's failure to detect a third party's omission.
- Definition of "modified." Limit mandatory labels to content that could reasonably mislead about a real person or event. Routine editing should be out of scope.
- Platform duties. Codify a notice-and-action model and an uploader self-declaration tool, and state expressly that there is no general obligation to monitor.
- Proportionality. Give small services longer phase-in periods and safe harbours for good-faith compliance.
- Standards. Prefer interoperable provenance standards over a bespoke Argentine label, so that local firms are not forced to build a market-specific system.
Bottom line
A transparency rule for synthetic media is a defensible policy, and Argentina is right to consider one. The risk is that platform obligations turn a disclosure law into a surveillance and takedown law. If deputies assign responsibility to the parties who generate and publish the content, keep platforms within the limits the Supreme Court drew in 2014, and confine mandatory labels to material that actually deceives, the bill can protect the public without chilling the open internet or the country's growing AI sector.