The heads of three AI companies asked for limits on AI, and the White House said no.

Not merely what can we do, but what ought we to do? — Pope Leo XIV, to the International Catholic Legislators Network, August 2026

Last week the people who build AI asked the world to slow them down, and within three days almost everyone with the power to do it declined. A tradition that has been asking about this for far longer walked into Congress the same week. Underneath both sits a question you will answer for your own organization long before Washington answers it: who decides where the line goes?

The story

The people building AI asked for limits, and everyone else said no

On Saturday, September 12, the chief executive of one of the three leading AI labs published an essay arguing that his own industry is moving too fast. Dario Amodei of Anthropic proposed three steps: let independent evaluators work inside the labs with the access an employee has, get the companies in democratic countries to agree on shared safety standards, then take that agreement to governments that are not democracies. He called it pacing rather than stopping.

Within a day, the people he competes with agreed. Elon Musk of xAI posted three words, "Dario is right." Sam Altman of OpenAI said he agreed the frontier needs pacing. Demis Hassabis of Google DeepMind joined them. On CBS Sunday Morning the next day, Amodei told Jo Ling Kent that the next conversation needs government in the room, and admitted he had not expected the pace to feel like this.

Then Monday came, and the answer was no.

President Donald Trump rejected the idea across a string of posts, writing that the danger of "AI taking over the World, destroying Humanity, and all other things bad, is a HOAX". He phoned Nvidia's chief executive Jensen Huang while Huang sat on a conference stage in Los Angeles, and used the call to wave the concerns away. The market took its own view the same day: Nvidia fell 3.4 percent and Micron 5.3 percent.

The pushback did not only come from Washington. David Sacks, who ran AI policy at the White House until March, told the labs to "stop pretending the motivation to slow down is purely altruistic" and to pace themselves rather than ask anyone's permission. China's foreign ministry, asked about it directly, would not engage with the substance, saying fear-mongering and vicious competition only get in the way of governing AI well.

So read the week again and watch where the responsibility travels. The labs asked for rules they cannot set alone. The White House said no rules are needed. The critic said the labs should simply act. The foreign ministry said the question itself is unhelpful. Four parties, and every one of them handed the job to someone else.

One step did happen. Amodei committed Anthropic to the outside evaluators on its own, without waiting for agreement.

That page takes an afternoon. It is the only step on the list that belongs to you.

More this week

A dozen bishops went to Congress to ask for AI safety rules

The day after the labs called for pacing, the U.S. Conference of Catholic Bishops created a task force on AI and human dignity and sent more than a dozen bishops to Capitol Hill to ask Congress for AI safety rules. They had met representatives of the technology companies first. Their stated ground was not risk forecasting but human dignity, drawing on the encyclical Pope Leo XIV published back in May.

Notice what that means. The bishops were not reacting to last week's news. They were already moving, on a document four months old, written before any lab chief changed his mind in public. A tradition that has spent centuries asking what tools do to people arrived at roughly the same place the labs did, from the opposite direction, and got there first.

There is a lesson in the timing for anyone who has been waiting for the technology to settle before forming a view. The people who had already done the thinking were the ones ready to act in the week it mattered.

The case for staying human while you use the tools

A group of writers set aside the governance argument entirely and asked a harder question: what do these tools do to understanding, memory, attention and friendship? "Offloading Ourselves" is the founding document of a new council on AI ethics, and it is the most forwardable thing published this season for anyone who has already read the productivity arguments and wants the deeper one.

Its sharpest moment is a quotation rather than an argument. Mark Zuckerberg has described people who do not wear AI glasses as being at a significant cognitive disadvantage. The writers simply ask what sort of advantage that is, if the thing it buys you is a mind that no longer does the work on its own. Hand this to the colleague who keeps saying the tools are just tools. It is the better conversation, and your team is already having it without you.

They published 8,913 fake news articles. Almost nobody read them.

Anthropic disclosed a commercial operation that built roughly seventy fake news sites and published at least 8,913 articles in about twenty languages — and then reported that most of it drew little real engagement from actual readers. The detail sits in the company's own threat report, alongside a network of more than 250 fake commenting accounts built to make the articles look discussed.

The number to know: 8,913 articles that almost nobody read. Generating text is now close to free, and attention never was. Whatever your marketing team is planning, publishing more is no longer a strategy, and it has not been for a while. The scarce thing is being worth reading, which is the one part the machine cannot do for you.

Also on the radar

• Gallup's running indicator shows 65 percent of employees say AI has helped their own productivity, while only 14 percent strongly agree it has changed how work gets done where they work. Individuals have adopted it. Organizations have not.
• Gallup also found that workers whose manager actively champions AI are far likelier to say it changed their work: 33 percent against 4 percent. The constraint is the manager, not the model.

The practice

Let your strengths sift what AI does for you

In a study of 758 consultants, the people who did best with AI were not the ones who handed it the most. They were the ones who knew what they did well and handed it the rest. The experiment, now published in Organization Science, found consultants using AI completed more work, faster. On one task chosen deliberately to sit outside what the tool does well, the people using AI were far likelier to get it wrong.

Same people, same tool, opposite result, depending entirely on whether the work played to the machine's strength or to theirs.

The MIT Sloan work is here, from March 2025, and its finding is that work has been shifting toward exactly those capabilities.

Here is the difference in one pair of prompts. Most people write the first.

Weak, because it hands over the judgment:

Strong, because it hands over the labor and keeps the judgment:

The second takes ninety seconds longer to write, and it is the whole practice. Name what you do best. Hand over the rest.

Your step this week: write down the one thing you do better than anyone on your team. Then look at what you gave AI in the last month and see whether you handed away that exact thing. Not anti-tech. Pro-flourishing.

Higher ground

On the Wednesday of that same week, Pope Leo XIV was working through a scheduled series on a document from the Second Vatican Council, not responding to any of this. He warned anyway about growing accustomed to handing algorithms decisions that belong to the human conscience alone.

The next step

You cannot set your organization's line until you know your own. That is what the AI Edge assessment is for: about four minutes, and it tells you how you are wired to work with these tools, so you stop handing over whatever the internet says to hand over.

Find your AI Edge →

Then hit reply and tell us what surprised you. We read every one.

Climbing with you,
Jeff and Jason