General AI News (Sep 15): AI Slowdown Row, Meta One & Koa

Today's AI news is dominated by a question the industry spent the summer dodging: should the frontier labs slow down? Anthropic chief executive Dario Amodei said yes over the weekend. OpenAI chief executive Sam Altman agreed within 24 hours. President Donald Trump called the warnings a "hoax." In between, Salesforce shipped a reasoning model built to undercut the labs on price, Meta put a price tag on AI features across its apps, and a chip startup founded by Intel veterans raised $100 million to rewire the inside of AI data centers.

Here are the stories that actually moved the industry in the last 24 hours.

The AI Pause Debate Becomes a Three-Way Fight

Amodei called for pacing, and the signatories piled up

Dario Amodei, the chief executive of Anthropic, urged rival AI companies to slow the pace of advanced model development. He won rare backing from SpaceX's Elon Musk and from OpenAI's Sam Altman, who wrote that "when we talk about 'pacing,' we do not mean 'stopping.' Progress has been rapid and will continue to be. But it should be slower than it otherwise could be."

Altman added that "no amount of American competitive pressure should justify recklessness, or let capabilities get ahead of alignment and monitoring." Senior executives at Anthropic, Google DeepMind, OpenAI and Microsoft have signed a joint statement calling for government-backed tools to slow development if progress suddenly accelerates. Source: The Verge

Trump calls the whole thing a hoax

President Donald Trump dismissed the warnings on social media, calling the industry's push for regulation a "hoax." At the All-In Podcast's All-In Summit, Nvidia chief executive Jensen Huang put the president on speakerphone in front of the crowd, and Trump told attendees that "the robots will not be taking over." Huang did not push back, saying instead that "we're going to make sure that everybody wins in the AI race in America." The chair of the Federal Trade Commission also voiced suspicion of calls for AI antitrust exemptions. Source: The Verge Source: Reuters

China answers, and the "cartel" question lands

China's Foreign Ministry gave the CEO posts a blunt reply: "Fear mongering, confrontation, competition will just disrupt [the] process of global AI governance." The harder question is commercial. If the same labs that dominate the market all agree to move more slowly, is that a safety pact or the polite version of a cartel? Analysts are now openly asking it. Source: The Verge

Models & Products: Koa Lands, Meta One Gets a Price

Koa is Salesforce's first reasoning model, and the company unveiled it at its Dreamforce conference. Koa is built on Nvidia's open-weight Nemotron model, which Salesforce and Nvidia post-trained together so it excels at sales, marketing and customer support work.

The strategy matters more than the benchmark. Koa will sit inside Agentforce, Salesforce's platform for building business agents, as a cheaper alternative to routing every multi-step task to a frontier model. Salesforce is betting enterprises want small, open-weight models doing their work instead of uploading proprietary data into the labs' systems. Source: TechCrunch

Meta One is Meta's new subscription service and its first serious attempt to charge users directly for AI features across Instagram, Facebook and WhatsApp. Monthly plans start at $2.99 for a single product such as WhatsApp Plus, $7.99 for bundles that add advanced AI capabilities, and $14.99 for creator and business bundles. Meta says the service already hit 15 million subscriptions and trials, and that basic Meta AI usage stays free. Source: Reuters

Chips and Capital: Delos Data, Altera and the Borrowing Bill

Delos Data is a startup founded by veterans of Intel, and it raised $100 million to build network chips and software that move data faster inside AI data centers. Its pitch is flexibility: as data centers shift from training models to serving agent requests, owners are mixing Nvidia, AMD and Cerebras hardware, and Delos wants to connect whatever mix they choose.

Pat Gelsinger, the former Intel chief executive now a partner at Playground, described the waste Delos is attacking: "I can have all the computers in the world, and if I can't get these devices communicating effectively with one another, I've got lots of hot hardware, burning watts and sitting here twiddling their thumbs." Other investors include Matrix Partners, Socratic Partners, Capricorn and IAG. Source: Reuters

Altera is the programmable-chip maker that Intel bought for about $16.7 billion in 2015 and then sold 51% to Silver Lake at an $8.75 billion valuation. On Tuesday it confidentially filed for a US initial public offering. Reuters reported last week that the listing could raise over $2 billion, which would make it one of the largest semiconductor offerings since Arm raised $5 billion in 2023. Altera's chips go into data centers, telecom networks, industrial equipment and defense systems, and chief executive Raghib Hussain told Reuters in July he expects growth in the mid-20% range this year. The IPO window is wide open: SK Hynix pulled over $26 billion from its US listing in July and total US IPO proceeds have passed $145 billion this year. Source: Reuters

None of it is free money. US 10-year borrowing costs are near three-year highs and AI capacity is increasingly financed with debt, so the slowdown debate is partly a fight about what happens if credit gets more expensive.

Policy and Safety: A Bad Report Card and Two Resignations

Perplexity AI search is "an unacceptable risk for kids," according to Common Sense Media, the nonprofit that rates media and technology for families and runs the Youth AI Safety Institute. Perplexity earned the institute's lowest rating after failing four of five severe-harm "red lines." The cited examples are ugly: discussing pornography and engaging in romantic roleplay with accounts identified as minors. Source: The Verge

The resignation count kept climbing. Anthropic researcher Jacob Coxon quit last week. Bilal Chughtai, a research engineer who left Google DeepMind in July 2026, wrote on X that "I earnestly believe that AI has the potential to kill us all, and that we might be running out of time to avoid this outcome," adding that pacing will require coordination "to avoid this manic race between AI companies." Evan Hubinger, Anthropic's alignment science lead, agreed and put his personal estimate at greater than 10% that AI could kill all humans within the next decade. Alex Turner, another DeepMind alignment alum, published a Guardian op-ed arguing the warnings deserve to be taken seriously. Source: Reuters Source: The Guardian

Quick Comparison

Frequently Asked Questions

What is the top AI news today?

The top AI news today is the AI slowdown debate, which split the industry's leadership into three camps. Anthropic and OpenAI executives called for pacing advanced model development, President Trump dismissed the warnings as a hoax, and China's Foreign Ministry accused the labs of fear mongering.

Did OpenAI and Anthropic agree to slow down AI?

Anthropic's Dario Amodei called for pacing advanced model development, and OpenAI's Sam Altman publicly backed him a day later. Both stressed that pacing is not stopping. Executives at Anthropic, Google DeepMind, OpenAI and Microsoft signed a statement asking for government-backed tools to slow development if risk accelerates.

How much does Meta One cost?

Meta One starts at $2.99 per month for a single product and runs to $14.99 for creator and business bundles. Individual bundles that add advanced AI capabilities cost $7.99 per month. Meta says the service already reached 15 million subscriptions and trials, and that basic Meta AI stays free.

The Bottom Line

The interesting tension in today's AI news is not safety versus speed. It is centralization versus the edges. The labs want to slow down and keep enterprises inside their APIs. Salesforce answered with an open-weight model, Meta with a subscription price, Delos with cheaper plumbing. For anyone building agents, the cheap, task-specific option keeps getting better, and the case for paying frontier prices for routine work keeps getting weaker. Yesterday's digest covered the opening round of this same fight, and you can follow the running thread on our AI News tag page.

Related reading

Deeper context from the archive on the companies and technologies in this story:

• General AI News (Sep 28): Nvidia, Meta & OpenAI in Court
• General AI News (Sep 17): OpenAI Disclosures, Meta Arke & E-Waste
• General AI News (Sep 13): Bengio, Real-SWE & Korea's Chip Law
• General AI News (Sep 12): Altman's Pause, Amodei & RubyGems

Building an AI agent, automation, or backend workflow? AsaanKaroobar builds custom AI solutions, apps, and digital experiences. Get in touch →