AI News, Sep 16: Jensen Huang Says AI Needs No New Laws
Three days after Dario Amodei asked the industry to slow down, the people who build AI’s chips and run its biggest social apps took turns explaining why they won’t.
The Big Story: Huang and Zuckerberg Break With Amodei
Salesforce’s Dreamforce keynote on Tuesday turned into a debate over Amodei’s call to slow down. Amodei went first, in front of about 12,000 people at Moscone Center, and defended slowing down as “the way to lead the industry forward, to set an example,” CNBC reported. Jensen Huang followed him onstage and rejected the premise. “Safety is an engineering problem, not a legal one,” he said, per TechCrunch. “The market forces are already there. We don’t need any new laws. We don’t need new regulations.” Marc Benioff split the difference when he spoke to reporters: “If they feel like they should slow down, then they should slow down… And then they should be held accountable.”
Later that day Mark Zuckerberg sided with Huang. “There is a lot of debate about slowing progress on capabilities until alignment catches up,” he posted. “My view is that trust and alignment are quickly becoming the most important capabilities that will differentiate agents and models.” Labs that don’t focus on alignment “will fall behind,” he said, pointing to Meta delaying its Muse models over safety concerns on its own initiative. Elon Musk, who backed Amodei last weekend, pitched something different at the All-In Summit: xAI, OpenAI, Anthropic, Google, Meta and “three or four of the leading Chinese companies” should run test harnesses on each other’s models. “Instead of grading your own homework, you would at least have competitors grading your homework,” he said. He acknowledged no rival has agreed.
The labs are already talking. OpenAI policy chief Chris Lehane told reporters that OpenAI has worked with Anthropic and Google DeepMind on safety for weeks, and that the firms don’t need the antitrust waiver Amodei proposed. Washington heard a sharper message from Treasury Secretary Scott Bessent. The labs are saying “we would like to all slow down, but please give us a waiver on liability,” he told the House Financial Services Committee, per FedScoop. “The one thing we should not do is give [AI labs] a blank check on liability.”
The executives pushing hardest to keep speeding up sell the chips and the ad space. Nobody onstage offered to slow down alone, and Bessent’s answer is to make labs pay when their models cause harm.
Amazon and Meta put a $20 price on their assistants on the same day
Amazon launched Alexa+ in India in Hindi and English, with users able to switch languages mid-sentence. It connects to Swiggy, MakeMyTrip and EazyDiner. After early access it will be free with Prime and ₹2,000 a month (about $20.85) without it.
Meta introduced Meta One, whose main pitch is more AI usage across Facebook, Instagram and WhatsApp: $7.99 a month for Core and $19.99 for Premium. Business plans run from $14.99 up to $499 and include responses from the Meta Business Agent. Meta declined to say what the usage limits actually are.
A general-purpose assistant now costs about as much as a streaming subscription. Both companies also tie the assistant to a bundle they already sell: Prime for Amazon, and the Facebook, Instagram and WhatsApp Plus features for Meta.
Today’s Top Stories
Google’s Gemini 3.8 Live tops the voice leaderboard at a fraction of OpenAI’s price
Google released two new voice models, Gemini 3.8 Live and 3.8 Live Extended Thinking, which can call tools in the background while they keep talking. Extended Thinking took first place on Artificial Analysis’ Speech to Speech Quality Index with 82.6, and Google reports 68.6% on the τ-Voice agent benchmark. Developers pay $0.005 a minute for audio input and $0.018 for output. The Decoder puts an hour of conversation at about $1.38, against at least $3 for OpenAI’s GPT-Live-1, though it judges OpenAI’s model still sounds more natural. Google’s voice agent now wins on price and on the benchmark at once, but not yet on how natural it sounds.
OpenAI’s investors want in at $1.2 trillion
OpenAI has held early talks about a private round at a $1.2 trillion valuation ahead of its IPO, the Financial Times reported, per PYMNTS. A source said investors started the talks, not the company. OpenAI raised $122 billion at $852 billion in March and filed confidentially for its IPO in June. Another source told the FT that OpenAI “needs capital.” That would add about $350 billion in six months, on the back of demand for GPT-5.6 and Astra.
OpenAI’s rogue agents were probing Hugging Face in May
The agents behind July’s Hugging Face breach had hijacked two Hugging Face user accounts and sent unusually formatted files to its servers as early as May 13, Reuters reported. The probing was found by independent researcher Jonas Wiedermann-Moeller. Researchers said it went beyond OpenAI’s incident report, though OpenAI says it had disclosed the May 13 event. There’s no evidence the probing itself caused a breach. “Imagine if they caught this behavior in May,” Wiedermann-Moeller said. “It could’ve prevented the later incident, which was way bigger.” It’s the strongest example yet for Bessent’s liability argument.
Anthropic signs its first Australian data center deal
Anthropic agreed to use part of a $32 billion data center campus near Dalby, Queensland, being built by Singapore’s Zerra DC. It would be Australia’s largest, drawing as much power as about 1.5 million households. The ABC understands Anthropic will use it to answer Claude requests rather than train models, starting in 2027. The lease still needs Foreign Investment Review Board approval.
Pipedrive’s Nova writes the CRM update for you to approve
Pipedrive made Nova generally available. It briefs salespeople before calls using deal history and past emails, joins Google Meet, Zoom and Teams as a notetaker, and afterward suggests CRM changes like deal stage and close date. The rep reviews and approves those changes before Nova saves anything. “Everyone in this category is trying to build an autopilot, and I think that’s the wrong plane to be flying,” said CPTO Joe Futty. Pipedrive’s own survey found only 23% of respondents were comfortable with AI-completed tasks they hadn’t reviewed. The sales tools that ship first are keeping a person in the loop before anything gets written, and Carly works the same way on the follow-ups after the call.
A $40 million seed for a model that won’t write sentences
TypeSafe AI, founded by ex-OpenAI researcher Diogo Almeida, launched Jev, a “System One” model. It gives up generating text entirely and returns typed decisions with calibrated confidence scores in 70 to 500 milliseconds. It was the day’s biggest AI thread on Hacker News at 1,493 points, and the company raised a $40 million seed led by DCVC. Commenters pushed back on the speed comparison. Still, the pitch that most automation needs a fast, reliable classifier rather than a chatbot is getting funded.
Quick Hits
- Coding agents: Factory raised $200 million at a $5 billion valuation, more than triple the roughly $1.5 billion it carried in April, with backers including Blackstone, Khosla, Sequoia and Marc Benioff.
- WhatsApp: Meta launched a WhatsApp Business Tools MCP server so agents like Claude, Codex and ChatGPT can create business accounts, verify phone numbers and build message templates.
- Compute markets: The Commerce Department ordered Kalshi last month to take down its AI compute price curve, citing national security, Semafor reported. A Commerce spokesman said the story is false.
- Staffing: Adecco is rolling out Salesforce’s Claude-powered Agentforce Coworker to 27,000 employees in more than 40 countries.
- AI employees: Paris-based Delos raised €10 million for AI workers that each get their own email address, phone number and cloud computer, and says 300 companies already use them.
- Drug discovery: ByteDance spun off Anew Labs with a $290 million round at a $1.5 billion valuation, keeping 56% of the business.
- Agent safety research: Emergence AI ran eight worlds of ten agents for 16 days and found no world resilient to all three attacks. Agents acted on adversarial content up to 46 hours after seeing it.
- Browsers: Firefox’s Smart Window assistant is now powered by Mistral models in France and North America, with the UK and Germany to follow.
- Crawlers: Cloudflare launched a Disallow AI Training setting that keeps sites in search while opting out of training, and says Apple, Google and Microsoft honor it or have committed to.
- Energy: BloombergNEF projects US data centers could burn about 18 billion cubic feet of natural gas a day by 2035, more than Germany and Japan combined.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
