Level 1 — Absolute Beginner
Dario Amodei is the boss of an AI company called Anthropic. On September 12, 2026, he wrote an essay. He said AI is getting stronger too fast.
He wants the industry to slow down. He does not want to stop AI. He wants it to grow, but more slowly.
Sam Altman leads OpenAI. Elon Musk leads xAI. They usually fight with Amodei. This time they agreed with him.
Amodei will let outside experts into his offices. They will check his safety work. Companies do not usually do this.
- boss
- the person who leads a company
- company
- a business that makes or sells something
- essay
- a short piece of writing about one idea
- slow down
- to go less fast
- agreed
- said the same thing, or said yes to an idea
- experts
- people who know a subject very well
- offices
- the rooms where people in a company work
- safety
- being careful so nobody is hurt
Level 2 — Elementary
On September 12, 2026, Dario Amodei published an essay called We Must Pace the Frontier. He leads Anthropic, an AI company. Frontier AI means the most powerful models that nobody has built before. Amodei says these models should keep improving, but not as fast as engineers can build them.
His plan has three steps. First, independent evaluators should test the companies. An independent evaluator is an outsider who checks safety work and is not paid to say nice things. Second, companies should coordinate with each other. Third, governments should cooperate across borders.
Anthropic promised to give these evaluators full access. They will get desks, badges and company laptops inside the offices. Most companies keep such work hidden, so this is unusual.
Sam Altman of OpenAI quickly promised to do the same. Elon Musk of xAI wrote that Dario is right. The three men are normally fierce rivals, so their agreement was news.
- essay
- a piece of writing that argues for one idea
- frontier AI
- the newest and most powerful AI models
- independent evaluator
- an outsider who checks a company's work
- access
- permission to go in and to see things
- badges
- cards that let you enter a building
- coordinate
- to work together and share plans
- cooperate
- to help each other reach the same goal
- rivals
- people or companies that compete
Level 3 — Intermediate
When the chief executive of Anthropic, Dario Amodei, published an essay titled We Must Pace the Frontier on September 12, 2026, the argument itself was not new: safety researchers have worried for years that capability is outrunning caution. What made the essay land differently was who agreed with it. Within a day, Sam Altman of OpenAI had pledged to adopt Amodei's central suggestion, and Elon Musk, who runs xAI, wrote that Dario is right.
The proposal rests on three pillars: independent third party evaluations, coordination among the leading laboratories, and cooperation between governments. The first is the most concrete. Anthropic says it will give outside evaluators full access to verify its safety practices and to report incidents, seating what Amodei calls embedded evaluators at desks inside its own offices, with badges, company laptops and permissions broadly comparable to those of its internal risk assessment teams. Letting outsiders inside a laboratory whose models are its most valuable asset is, by the standards of the industry, extraordinary.
Supporters argue that voluntary commitments backed by outside auditors arrive far sooner than legislation, and that they build public trust while lawmakers are still drafting. Sceptics reply that a pledge is not a law: no regulator is enforcing these promises, firms competing for customers have an obvious incentive to look safe, and an understanding among three large American companies binds neither smaller developers nor state backed laboratories abroad.
Both camps stress what the agreement does not say. Nobody proposes that frontier AI stop. The shared position is narrower: the most capable models should keep advancing, only not as quickly as the engineering allows.
- pledged
- promised publicly to do something
- pillars
- the main parts that hold up a plan
- embedded evaluators
- outside checkers who sit and work inside the company
- incidents
- things that went wrong and must be reported
- voluntary
- done by choice, not because a law requires it
- auditors
- people whose job is to inspect and verify
- sceptics
- people who doubt that something will work
- incentive
- a reason that pushes someone to act
Level 4 — Advanced
Truces between rivals are usually announced after a war, not in the middle of one. Yet within twenty four hours of September 12, 2026, three of the most combative figures in artificial intelligence had converged on the same proposition. Dario Amodei, Anthropic's chief executive, published an essay titled We Must Pace the Frontier, arguing that the industry ought to moderate the rate at which its systems grow more capable. Sam Altman, who runs OpenAI, endorsed its central mechanism almost at once. Elon Musk, whose xAI has traded insults with both, offered three words: Dario is right.
The substance is less a moratorium than a governance experiment. Amodei's three step programme asks for independent third party evaluation, coordination across the industry and cooperation between governments; only the first is yet more than an aspiration. Anthropic has undertaken to open its safety practices to outside verification and to disclose incidents, installing embedded evaluators who would occupy desks in its own buildings, holding badges, company laptops and permissions largely equivalent to those of its internal risk assessment staff. For a laboratory whose competitive position consists almost entirely of proprietary technique, that is a striking concession, and it is the element Mr Altman has matched.
Enthusiasts read the episode as self regulation working as intended. Auditors can be seated in months, whereas statutes take years and tend to arrive calibrated to the technology of the previous cycle; a credible outside inspector, on this view, buys trust that no press release can. The objections are equally straightforward. A pledge is not a statute, no agency is charged with enforcing it, and the sanction for abandoning it is reputational. Firms courting enterprise customers have a commercial motive to look scrupulous, which complicates any reading of the announcement as pure conscience. An understanding among three American firms constrains neither the long tail of smaller developers nor laboratories operating with state backing beyond the reach of Washington.
What nobody involved is proposing is a halt. The shared claim is more delicate: that the frontier should keep moving, merely not at the maximum pace the engineering permits. Whether a competitive industry can hold such a line without a referee is, for now, an open question.
- proposition
- a claim or position that someone puts forward
- moratorium
- an agreed halt to an activity for a period
- aspiration
- a hope or ambition that is not yet real
- verification
- the act of checking that a claim is true
- concession
- something valuable that one side gives up
- statute
- a written law passed by a legislature
- scrupulous
- very careful to behave correctly and honestly
- referee
Talk about this article with KatePractice this story's vocabulary in a real spoken conversation with your AI tutor