Preservation or Extinction? Inside the Algorithmic Threat to Indigenous Languages in Brazil

Brazil’s latest technological obsession is not a fintech or an AI startup promising to save the Amazon. It is an app—BILingo—marketed by universities and cultural agencies as an “inclusive” language-learning platform capable of teaching Bororo, Makurapi, and other Indigenous languages to the masses. Newspapers call it “Duolingo indígena.” Ministries praise its innovative spirit. Academics hail it as a breakthrough in “digital preservation.”

It is, at first glance, the kind of frictionless feel-good story that populates the innovation section of national newspapers. A shiny product. A promise of accessibility. An apparent antidote to the erasure of Indigenous languages.

And yet, when you scratch the surface, BILingo reveals something far more uncomfortable: the re-emergence of old colonial patterns—this time mediated by algorithms, datasets, and servers hosted thousands of kilometers away from the communities whose knowledge sustains them.

This is what Indigenous thinkers like Anápuàka Tupinambá Hãhãhãe have explicitly named: genocídio algorítmico. Not the theatrical apocalypse imagined by Silicon Valley critics, but a tangible, bureaucratic, quietly lethal mechanism of cultural displacement. It is the moment when the machine begins to speak in place of the elders.

At the center of the controversy is a simple, devastating question:
Who controls the knowledge that feeds the algorithm?

BILingo is being developed by the University of São Paulo and the University of Tübingen, with the support of state agencies and, tellingly, missionary organizations. Its linguistic models are trained on voices, stories, ritual speech, archive material, and translations—some of them biblical translations produced through missionary interference in the 20th century. These recordings and texts were never meant to be “datasets.” They are the backbone of living cultural systems.

But for the algorithm, they are just tokens.

For AI engineers, they are just inputs.

For institutions hungry for innovation headlines, they are assets.

And in none of these institutional layers do Indigenous communities hold structural authority. There is symbolic consultation—yes. Occasional participation of youth—also yes. But protagonismo indígena, in the full sense of sovereignty over data, design, deployment, licensing, and refusal? That is absent.

This is where the colonial logic becomes unmistakable.
The app is built about Indigenous Peoples.
It is justified for Indigenous Peoples.
But it is not governed by Indigenous Peoples.

Brazil knows this pattern well: it is the same choreography that shaped indigenismo for decades—public institutions that speak in the name of Indigenous communities while retaining all decision-making power. Even when Indigenous individuals occupy high-level positions, the structural authority remains with the same non-Indigenous bureaucracies and academic elites.

Digital technology did not disrupt this logic. It streamlined it.

And this is where the notion of genocídio algorítmico becomes technically precise. When AI reconfigures a language—removing its rhythm, its cosmology, its ceremonial grounding—the result is not a neutral simplification. It is the production of a parallel version of that language, one that may eventually become more “authoritative” to outsiders, to students, even to future generations of Indigenous youth.

In other words, the app becomes the archive.
The algorithm becomes the tutor.
The machine becomes the reference.
And the elder becomes obsolete.

This is not preservation. It is substitution.

And once a language is standardized, tokenized, and encoded, it becomes open to an even more disturbing possibility already emerging abroad: blockchain commodification. Companies have begun minting Indigenous languages and cultural fragments as NFTs, selling “immersive experiences” to buyers with no accountability to the communities whose identities are being monetized. An app like BILingo—unregulated, ungoverned, and data-hungry—paves the road for similar exploitation in Brazil.

There is a profoundly colonial fantasy at play here: that technology can “save” Indigenous languages without Indigenous governance. That modernity will repair what modernity destroyed. That the solution to epistemicide is an app.

But Indigenous communities are offering a radically different framework—Protocolos Comunitários de Ética, Dados e IA, grounded in UNDRIP, the OIT 169 Convention, and Indigenous epistemologies. These protocols are not bureaucratic hurdles; they are survival infrastructures. They establish that:

• data must be collectively governed;
• communities must retain veto power;
• elders must be authors, not sources;
• language cannot be extracted without ritual, context, and purpose;
• AI cannot overwrite cosmology with convenience.

The question is not whether technology should engage with Indigenous cultures. The question is under what terms, whose governance, and with what limits.

Because without Indigenous authority over the digital, every AI tool becomes just another frontier of extraction—the cross and the sword replaced by the app and the algorithm.

Brazil does not need an “Indigenous Duolingo.”
It needs Indigenous digital sovereignty.
Anything less is not innovation. It is continuity.