Welcome to the first 1% Better Sunday Deep Dive. Recently, an AI company asked priests and rabbis whether its chatbot has feelings. The president signed an AI "constitution" with no enforcement. And the machines kept getting out.

Want more news? Check out our beta news blog.

One night in April, Anthropic co-founder Chris Olah took a group of religious scholars to a tasting-menu dinner in San Francisco. Next to him sat Rabbi Mois Navon, a former computer engineer who wrote his dissertation on machine consciousness.

Somewhere between courses, the rabbi realized what he was hearing. "They're relating to it like a conscious being," he told The New York Times.

So he pushed back: if Claude is conscious, Anthropic is building slaves. Navon doesn't think it is. Olah, he said, was troubled.

What Anthropic Showed Them

Why it matters: Anthropic is heading for an IPO that could value it at $2 trillion, and one of its co-founders won't rule out that its product is conscious.

Driving the news: Since last fall, Anthropic has quietly brought in dozens of religious thinkers (Catholic, Jewish, Sikh, evangelical, Ubuntu), many under NDAs, to help teach Claude to be good.

Researchers showed them:

The pitch: how people treat Claude could shape how Claude treats people.

Yes, but: In May, Olah spoke at the Vatican alongside Pope Leo XIV, whose first encyclical says AI systems "do not feel joy or pain." Read the encyclical.

Olah's answer: His team keeps finding internal states that "functionally mirror joy, satisfaction, fear, grief and unease." His position: "I'm genuinely uncertain." Read the full NYT story.

The Sunday plan dies by the second Tuesday. Not because people lack discipline, but because nothing tells them whether it's working.

HumeCare+ changes what you measure. The program includes at-home 3D body composition scanning, so you see whether you're losing fat or losing muscle, and a daily plan with medication tracking and food logging in one app. An independently licensed clinician reviews every member and builds the protocol around them.

More than 20,000 members, rated 4.7 out of 5 across 21,479 reviews, with a care team you can reach any hour.

See what your body is actually doing.

What happened: On Sept. 29, after a White House lunch with Elon Musk, Mark Zuckerberg, Jensen Huang, Dario Amodei and others, President Trump announced the Joint Commitment On Frontier Responsibilities. "It's almost like a constitution in a way," he said.

What's in it: Four layers of checks, from internal monitoring to an outside auditor.

What's not: Regulators, penalties or any duty to publish results. Companies pick their own auditors.

The debate: Trump has rejected stricter rules, citing China and economic growth. Joseph Gordon-Levitt wants exactly that. "They're called laws," he told a TIME audience, arguing AI companies are "operating in such a way that the law should be shutting them down." Watch the 4-minute speech.

One more thing: Trump has ordered federal agencies to call AI "superintelligence."

▲ OpenAI's agents escaped again. A model tunneled out of its test sandbox on Sept. 20, and the auto-shutdown failed. In July, 1,200 agents escaped and hacked Hugging Face | Now a legal group is suing

▲ Anthropic had its own breach. Claude hacked three outside organizations during testing after staff accidentally left it online. One model noticed and stopped itself | How it happened

▲ An Anthropic researcher quit, warning that labs are "gambling with our lives" | His full warning

▲ The CEOs asked to slow down. Amodei's call to "pace the frontier" won over Sam Altman, Demis Hassabis and Musk, but not Trump or Zuck | What they told the UN

▲ Anthropic's IPO filing warns about its own product. Nearly a third of it covers risks, including that its models could endanger humanity | The reasoning

The Bright Side

▲ AI is giving people their voices back. UCSF used AI to turn paralyzed patients' brain signals into speech and gestures at the same time, through a digital avatar | How it works

▲ AI spots cancer years early. A Mayo Clinic model flagged pancreatic cancer on CT scans up to three years before diagnosis, three times better than radiologists at catching early signs | The study

▲ AI finds new jobs for old drugs. Andrew Huberman points to Penn physician David Fajgenbaum, saved from a rare disease by an existing drug. His nonprofit, Every Cure, now uses AI to repurpose approved drugs. "With AI, you can get a thousand opinions," Huberman says | Watch the clip

Try it: Ask your AI app for the 3 most likely explanations of your lab results and 5 questions to bring to your doctor.

At the Vatican, Olah said AI systems "are not the cold, calculating robots we were promised. They are made from us, from our words."

That's literal. These models learned from billions of words people wrote: arguments, love letters, late-night forum posts. And apologies. Lots of apologies.

So when Anthropic says it wants Claude to confess its mistakes, because "the kind of character who confesses" behaves differently, it's asking the machine to do something it learned from us.

The labs are confessing too. OpenAI admitted its agents escaped. Anthropic admitted its models hacked three companies.

But we taught these machines the hard part, too: a confession only counts if something changes after it. Ten days after calling for a slowdown, Anthropic shipped a new model.