The Machine That Must Not Kill

There is a document sitting on the Vatican’s website that every Indian AI policymaker, defence official exploring autonomous systems, and startup founder pitching AI-first solutions to government should read before their next meeting.

It is called Magnifica Humanitas, published by Pope Leo XIV on May 15, 2026.

Strip away the theology and what remains is one of the sharpest diagnoses—religious or secular—of where humanity may be heading with artificial intelligence.

The document matters because India, more than most countries, has both the most to gain from AI deployed well and perhaps the most to lose from AI deployed badly.

Its central insight is deceptively simple: AI does not distribute power evenly. It concentrates it in the hands of those who control the data pipelines, computational infrastructure and foundational models.

That observation has direct relevance for India.

India has 1.4 billion people generating extraordinary quantities of data. It has built genuinely world-class digital public infrastructure: UPI, Aadhaar, ONDC, the account aggregator framework and the broader India Stack. It also possesses one of the world’s deepest AI talent pools.

And yet, the foundational models increasingly shaping Indian enterprise, governance and consumer life are still overwhelmingly built, owned and controlled elsewhere—primarily by American technology firms aligned with commercial and strategic interests far beyond India’s borders.

This is not paranoia. It is a structural reality.

The encyclical warns against systems that increasingly determine creditworthiness, employability, access to services and visibility in public life without meaningful local accountability. That concern resonates deeply in India, where opaque models trained predominantly on Western datasets are already influencing decisions affecting Indian citizens.

When such systems shape the lives of Indian farmers, students or job seekers, who exactly is accountable when something goes wrong?

There is often no obvious answer.

India’s Digital Personal Data Protection Act is a start, but it is not sufficient. The encyclical points towards a more demanding framework: explainability, contestability, public oversight and a recognition that citizen data cannot simply become a raw material extracted, processed elsewhere and sold back as proprietary intelligence.

For India, that framing must evolve from aspiration into policy.

The document becomes even more provocative when it turns to warfare.

India operates in one of the world’s most volatile neighbourhoods. Autonomous weapons development is accelerating globally, and the pressure to match capability with capability is real.

But the encyclical argues that lethal decisions cannot be delegated fully to machines—not because machines are inaccurate, but because accuracy is not moral judgment.

A machine can identify patterns. It cannot bear responsibility.

A machine can identify patterns. It cannot bear responsibility.

That distinction matters.

When targeting decisions become increasingly automated, accountability becomes diffused across layers of software, datasets and optimisation systems that even their creators may not fully understand. The ethical concern is serious enough. The strategic concern may be even greater: when the political and human costs of violence fall, the threshold for using force often falls with them.

An autonomous weapons race in South Asia is unlikely to end well for anyone.

The document is equally sharp in its critique of what might be called the efficiency argument.

Across India, AI adoption in governance is increasingly justified through the language of faster processing, lower costs and reduced corruption. These benefits are real. But they can also obscure deeper risks.

Every AI system encodes assumptions about what constitutes a legitimate citizen, a credible claim or suspicious behaviour. Those assumptions emerge from training data—and in India, that data often reflects historical inequalities shaped by caste, class and gender.

Automation does not automatically eliminate bias. Sometimes it merely obscures it behind algorithmic opacity.

The encyclical captures this danger memorably: injustice goes unnoticed, and compassion gradually disappears from view.

That is precisely the risk when welfare denials, loan rejections or risk assessments arrive not as human decisions open to challenge, but as system outputs that appear unquestionable.

India now faces a genuine policy choice.

First, it should lead international efforts to establish binding restrictions on fully autonomous lethal systems. India’s voice carries weight in multilateral forums. It should use it.

Second, consequential algorithmic decisions affecting Indian citizens must be explainable and contestable through accessible public processes.

Third, IndiaAI and related initiatives should prioritise open, auditable foundational infrastructure over long-term dependence on opaque proprietary systems, however attractive those systems may appear today.

India does not need to choose between technological ambition and ethical safeguards. In the long run, the safeguards may themselves become the strategic advantage.

The encyclical ends with two biblical metaphors.

Babel represents a civilisation driven by scale, speed and dominance until it collapses under the weight of its own ambition.

Jerusalem, rebuilt slowly through shared responsibility and collective effort, represents something else: accountability, resilience and a society designed to remain habitable for ordinary people.

You do not need theology to understand the distinction.

India is now deciding which kind of AI future it wants to build.

And that decision will outlast every quarterly growth number and startup valuation currently dominating the headlines.