📜 Pope Leo XIV publishes the first encyclical on artificial intelligence

Pope Leo XIV published the first encyclical on artificial intelligence in history. It's called Magnifica Humanitas — "Magnificent Humanity" in Latin — and was signed exactly 135 years after Rerum Novarum, the founding text of the Church's social doctrine on workers' rights. The parallel is no accident: for this Pope, AI is the social and moral challenge of our era. Meanwhile, Meta recorded its employees to train the AI that could replace them, and a researcher discovered a hack as elegant as it is terrifying. The same week, three different ways of asking what it means to be human when machines learn.

📜 Pope Leo XIV publishes the first encyclical on artificial intelligence

On Sunday, May 25, Pope Leo XIV published Magnifica Humanitas, the first pontifical document dedicated entirely to artificial intelligence. The title — "Magnificent Humanity" in Latin — is no accident. Neither is the date: it was signed on May 15, 135 years after Leo XIII's Rerum Novarum, the founding text of the Church's social doctrine on workers' rights during the Industrial Revolution. The parallel is deliberate: AI is, for this Pope, the social and moral challenge of our era.

The most interesting part isn't just the content, but how it was presented. In the Synod Hall, alongside cardinals and theologians, was Christopher Olah, co-founder of Anthropic and head of its interpretability team. Not because of his position, but because research into model transparency is exactly the kind of work that concerns the Vatican.

The encyclical addresses three main axes: human dignity facing systems that make decisions for us, redistribution of value generated by automation, and the need for algorithmic transparency as a condition for a just society. Nothing about banning AI. Quite the opposite: the document calls on the Church to actively participate in the development of technology, not to observe from a distance.

From ethics at the Vatican to ethics in Meta's hallways: the same week showed that not all companies interpret responsibility the same way.

Meta records its employees to train its AI

Meta has confirmed it's recording its employees — meetings, conversations, interactions — to train its AI systems. The official justification: improving models' ability to understand natural conversations. The less comfortable reality: the same employees whose data is used for training could be training the AI that eventually replaces them.

And meanwhile, a security researcher discovered a method to manipulate language models using "ghost tokens": invisible sequences that, embedded in a prompt, alter the model's behavior without the user knowing. It's not a bug — it's a property of how transformers work. And there's no easy fix.

An encyclical calling for participation, a company recording its own employees, and a vulnerability with no easy fix. The question of what it means to be human when machines learn doesn't have one answer — but more and more institutions, from the Vatican to security labs, are trying to find it. And that, in itself, is progress.

— Max