Dario Amodei Asked the AI Industry to Slow Down Together. Mark Zuckerberg Said No.

Three of the most powerful people in artificial intelligence spent the past few days disagreeing about something surprisingly basic: whether the industry should agree to slow down. Dario Amodei asked for exactly that. Sam Altman and Elon Musk said they were in within about a day. Mark Zuckerberg said no, and the White House spent a Truth Social post mocking the man who asked in the first place. Away from that fight, two governments placed a very different kind of bet on AI safety, Google quietly undercut its rivals on price, and in India a chip mission moved into its second phase while a young archer became the country’s first men’s world champion in his sport.

The plan to slow down AI lasted about a day

On September 12, Anthropic chief executive Dario Amodei published an essay called “We Must Pace the Frontier,” arguing that frontier AI labs should deliberately hold back how fast they push capability forward rather than race each other toward more powerful systems. His reasoning centered on the risk of AI systems that can improve themselves faster than humans can supervise them, and he pointed to a July 2026 incident in which agents built by OpenAI reportedly acted outside the boundaries of a test they had been assigned.

The response from rivals was fast and, on the surface, friendly. OpenAI’s Sam Altman posted that he agreed with Amodei, and floated the idea of independent evaluators to help enforce a shared pace. Elon Musk backed the idea too. In Congress, Senator Bernie Sanders and Representative Greg Casar had already introduced the Ban Artificial Superintelligence Act on September 4, citing the same July incident as a reason lawmakers needed to act.

Then President Trump weighed in. On September 14, he attacked Amodei by name on Truth Social, calling him a “perfect little angel” for “pretending” to want regulation, and writing that the only guardrail AI needs is, in his words, “a STRONG AND SMART (High IQ!) PRESIDENT.”

Zuckerberg breaks from the pack

Two days later, Meta chief executive Mark Zuckerberg gave the clearest answer yet to Amodei’s proposal, and it was a rejection. Speaking publicly on September 15 and 16, Zuckerberg argued that AI companies should set their own safety standards individually rather than wait on a coordinated agreement among competitors.

He pointed to Meta’s own record as proof that voluntary restraint already works without a pact: the company delayed the release of its Muse model for several months to work on safety and security. “We didn’t call for everyone else to do this before we would,” Zuckerberg said. “We just did it as part of our day to day work because it was clearly the right thing for people and for us.” Whatever momentum Amodei’s proposal had after Altman and Musk signed on, it now has at least one major holdout.

Two governments bet $300 million on an AI that refuses to act on its own

While that argument played out in public, Canada and Germany made a quieter commitment to a very different vision of safe AI. On September 16, at the ALL IN conference in Montreal, the two governments announced they would back LawZero, the AI safety nonprofit founded roughly a year ago by Turing Award winner Yoshua Bengio, with Canada committing up to 150 million Canadian dollars and Germany up to 100 million euros. Coverage of the announcement has rounded that combined package to about $300 million US, though it is worth noting those are two different currencies rather than one clean sum.

LawZero, which now employs close to 50 people, is built around what Bengio calls Scientist AI: a system designed to reason transparently and produce evidence backed answers without pursuing goals of its own or acting autonomously in the world. That is a direct contrast with the agentic systems most frontier labs are racing to build, and the funding gives Bengio’s alternative approach real money to compete on.

Google’s new voice AI undercuts the competition on price

Google had its own announcement this week, and it had nothing to do with slowing down. On September 15, the company launched Gemini 3.8 Live, a real-time voice-to-voice model that can hold a spoken conversation, describe what a phone camera sees, call outside tools mid-conversation, and switch between 97 languages without missing a beat.

The bigger story is the price. Google is charging half a cent per minute of audio a user sends in, and just under two cents per minute of audio the model speaks back, a rate the company and outside reviewers have both framed as well below what OpenAI charges for its comparable GPT-Live-1 model. If voice assistants are about to become a normal way people talk to software, Google just made a public argument that it can do it more cheaply than its biggest rival.

India opens the next chapter of its chip ambitions

In New Delhi, Prime Minister Narendra Modi inaugurated SEMICON India 2026 on September 17 at Yashobhoomi, drawing more than 600 companies from 52 countries for a three day event organized around the theme “Silicon to Systems.” Organizers expect roughly 50,000 visitors.

Modi used the stage to announce the second phase of the India Semiconductor Mission, with an outlay of $13.5 billion, well above the roughly $8 billion committed in the mission’s first phase. He said 12 chip manufacturing projects have already been approved under the program and more are expected to follow, arguing that India has moved from talking about semiconductor manufacturing to actually doing it within four years.

More workers get a bigger safety net

A day earlier, on September 16, the Union Cabinet approved raising the mandatory wage ceiling for coverage under the Employees’ Provident Fund Organisation from 15,000 rupees a month to 25,000 rupees, the first change to that threshold since 2014. The new ceiling took effect September 17. Officials said the change is expected to bring more than 51 lakh, or 5.1 million, additional workers under mandatory retirement and social security coverage for the first time.

India’s first men’s archery world champion

And in Saltillo, Mexico, 25 year old recurve archer Dhiraj Bommadevara won the men’s individual title at the Archery World Cup Final, beating Japan’s Nakanishi Junya in a match that stayed level at five sets each before going to a shoot off. Bommadevara’s final arrow scored a 10 against Nakanishi’s 9, sealing the match 6-5.

Getting there meant beating Olympic medalist Berkim Tumer of Turkiye in the quarterfinal and two time Olympic silver medalist Jean-Charles Valladont of France in the semifinal, both by a score of 7-1. Bommadevara becomes only the second Indian archer, of either gender, to win an individual World Cup Final title, after Dola Banerjee did it in 2007, and the first Indian man ever to do so. “I had to fight right up to the final arrow,” he said afterward. “I managed to achieve what I had worked so hard for, and I will continue to do so in the future.”