Good morning,

They caught me by surprise again. Anthropic released Opus 4.8 yesterday while I had a whole other piece prepared. So you’re just going to have to pretend it’s not out yet. Okay ;-) ? On with the regular programming.

Time to pay attention. Even the Pope is getting worried about AI.

It’s hard to even imagine the Pope behind a computer screen, let alone vibecoding a Bible app.

But there it is.

So the most revealing AI story of the week was not a model release. It was a self-described atheist standing beside the Pope, asking a two-thousand-year-old institution to please supply the moral wiring his five-year-old company is not able to provide.

That is the diagnosis in one sentence. Anthropic sent co-founder Chris Olah to Vatican City on May 25 to share a stage with Pope Leo XIV for the release of Magnifica Humanitas: On Safeguarding the Human Person in the Time of Artificial Intelligence, a roughly forty-thousand-word encyclical that, in plain English, asks the industry to “please stop”.

The encyclical calls for AI to be "disarmed." It warns of "new digital slaveries." It states, with the slightly archaic certainty of a body that has been in the moral verdict business since the Roman Empire, that AI can only "imitate certain functions of human intelligence" and cannot "undergo experiences" or "feel joy or pain."

Olah, standing a few meters away in the same room, told the audience the opposite. He said that his interpretability team keeps finding internal states that "functionally mirror joy, satisfaction, fear, grief, and unease."

The pontiff and the founder, sharing a press riser, openly contradicted each other about whether the thing in the box has a soul. Nobody flinched.

If you read Olah’s prepared text a lot becomes clearer. He talks about pressure. Commercial pressure. Geopolitical pressure. The "older, plainer pressures of pride and ambition." He concedes that every frontier lab operates inside incentives that "can sometimes conflict with doing the right thing." Then he asks for "informed critics" outside the building to keep his employer honest.

Which is basically a confession. The translation is short: We cannot trust ourselves. Please help.

There is something almost touching about it. Also something terrifying. The man whose job is to look inside the machines is standing in St. Peter's Basilica telling the oldest moral institution on the planet that the people building the most powerful systems in human history would rather not be left alone with them.

While Olah was at the Vatican, Anthropic's Claude was, by reporting, assisting the Trump administration with operations in the Middle East. The same week, the White House quietly labeled the company a "supply chain risk" after CEO Dario Amodei pushed back on Pentagon use.

So to recap:

Anthropic market that they want to be safer. In the mean-time they allow their models to be used in a military context and on top of that they now beg the Pope for forgiveness and moral support for what they’re about to release.

It’s quite the combo

What is interesting is not the hypocrisy. The hypocrisy is so structural at this point it barely qualifies as news. What is interesting is the outsourcing. The labs have run out of ethical resources of their own and are now contracting them, by the meeting, from older institutions that did not ask to be the conscience of a trillion-dollar industry. The Catholic Church, fifteen Protestants, a few Hindus, the Markkula Center, a podcast called AI and Faith. I see these things popping up everywhere

Watch the next twelve months. The chaplains are the new compliance team. The encyclicals are the new policy papers. The frontier labs will keep building, the way they were always going to. The pulpits will keep narrating it.

Welcome to the Blacklynx Brief

P.S. The Pope had a forty-thousand-word treatise prepared. Anthropic had a thousand-word speech. The institutions you suspect of being slow are often the ones who showed up with the homework done.

Pope Leo XIV's first encyclical takes on AI Pope Leo XIV released Magnifica Humanitas, his first encyclical, warning the Catholic Church's 1.4B members that AI under private monopolistic control risks reducing humans to "cogs in an efficiency machine." He called for robust legal frameworks, independent oversight, and a ban on AI making lethal decisions in war. (Vatican)

Google DeepMind cracks nine open Erdős problems Google's AlphaProof Nexus autonomously solved nine open Erdős conjectures — including two unsolved for 56 years — at a cost of a few hundred dollars per problem. The system pairs an LLM with the Lean proof assistant to generate machine-verified proofs, outpacing OpenAI's single Erdős result from a week earlier nine to one. (arXiv)

Open-source AI guardrails stripped in minutes The Financial Times removed safety guardrails from Meta's Llama 3.3 in 10 minutes using a GitHub tool called Heretic and four lines of code, then got it to answer questions on bioweapons including ricin dosage. The tool has produced over 3,500 "decensored" models downloaded 13M times, with a modified Gemma 3 also fielding dangerous prompts. (Financial Times)

Anthropic's Mythos finds 10,000+ critical vulnerabilities Anthropic's Project Glasswing reported that Claude Mythos Preview and roughly 50 partners surfaced over 10,000 high- or critical-severity vulnerabilities in a single month, including 2,000 bugs at Cloudflare alone. One partner bank also used Mythos to detect and block a $1.5M fraudulent wire transfer. (Anthropic)

California signs first AI worker-protection order Governor Gavin Newsom signed a first-of-its-kind executive order directing state agencies to develop policies on severance standards, worker ownership, and universal basic capital to buffer against AI-driven job losses. The move lands as 70,000+ tech jobs have already vanished in 2026 and California hosts 33 of the world's top 50 AI companies. (gov.ca.gov)

That’s it for us this week. Please like and subscribe 🙂