AI News, Sep 15: Trump Calls AI Concerns a Hoax
On Saturday Dario Amodei wrote that AI companies should slow the pace of development. On Monday the President answered him by name, and Congress made clear it won’t answer before the election.
The Big Story: Trump Named Amodei and Called the Warnings a Hoax
Donald Trump spent Monday on Truth Social rejecting guardrails on AI. His administration, he wrote, “has stopped AI ‘people’ from doing bad, or potentially bad, ‘things,’ like Dario (Anthropic!), who is now pretending to be a ‘perfect little angel.’” He added that “We already have tremendous CRIMINAL and REGULATORY power over these companies!” and that the only control AI needs is “a STRONG AND SMART (High IQ!) PRESIDENT,” NBC News reported.
Then he called Jensen Huang, who was onstage at the All-In Summit in Los Angeles and put the president on speaker for the room. “It could also be China. And we’re not going to let that happen. It’s a hoax,” Trump said. Huang replied: “You’re right. We’re not going to let that happen, sir.” Vice President JD Vance went after the motive. Frontier AI companies “begging the government to regulate them,” he said, “feels a little bit to me like a bit of a Trojan horse.”
Congress won’t settle it either. The Senate has three AI bills and no timetable for a vote on any of them, including John Kennedy’s kill switch, arriving Wednesday, and Bernie Sanders’ ban on artificial superintelligence. Majority Leader John Thune has said the votes aren’t there for the broader bill he’s negotiating with Amy Klobuchar. The House adjourns at the end of the week and won’t return until after the election. “So you think we’re going to get AI done?” asked Sen. Rick Scott.
The labs asked Washington to set the pace. The White House says there’s nothing to pace, and the House is leaving town.
Superhuman, Apple and Amazon are wiring meetings, email and calendar into one assistant
Superhuman bought Fathom, the AI notetaker with over 400,000 monthly active users, valued at $94 million in 2024 per PitchBook. Superhuman already had an email client, a docs app, a calendar and a newly launched agent builder. With a notetaker in the middle, it can draft emails, schedule follow-up meetings and set agents to work on meeting data. “If we were to operate Fathom as a stand-alone company, we would have to build a lot of things that are already in the Superhuman platform,” said Fathom CEO Richard White.
Apple came at the same job from the operating system. TechCrunch frames macOS 27’s new Siri as a rival to AI productivity apps like Highlight, Caddy and Littlebird, and the iPhone version is built on Google’s Gemini models. Amazon’s Monday AWS roundup listed last week’s general release of the Amazon Quick desktop app, which stays connected to your calendar, email and business apps in the background.
A notetaker, a screen-aware assistant and a work agent used to be separate purchases. Within a week, each became a feature of something bigger.
Today’s Top Stories
OpenAI is reportedly buying the team behind Apple’s Portrait Mode
OpenAI is buying smartphone camera maker Glass Imaging in a deal worth over $300 million, according to The Wall Street Journal. Founders Ziv Attar and Tom Bishop previously led the Apple team that built Portrait Mode. OpenAI already paid $6.5 billion for Jony Ive’s device startup io. A company building its own hardware needs a camera that holds up next to an iPhone, and this buys the people who built the iPhone’s.
Contractors are reading real ChatGPT conversations
404 Media reported that contract workers review real ChatGPT chats in a program called Project Lily, recruited through Crossing Hurdles and paid through Mercor, per The Decoder. Some of the reviewed conversations included users explicitly asking ChatGPT to keep the contents private. Anthropic confirmed it also uses human reviewers on Claude responses. Turning off “Improve the model for everyone” keeps chats out of training and away from human readers, but the setting is on by default and isn’t retroactive.
Google now lets its engineers use Claude
Google had barred employees from outside coding tools like Claude Code, with only a few teams allowed to use Claude. Now engineers can use Anthropic’s Opus 5 through Antigravity, Google’s in-house platform, under per-user quotas, Seoul Economic Daily reported, citing Business Insider. “Gemini remains the flagship and foundational model for internal development,” Google said. The company with the most riding on Gemini just decided its engineers move faster with a rival’s model on a quota.
Salesforce built its own CRM model on Nvidia’s open weights
Salesforce announced Koa, post-trained from Nvidia’s Nemotron 3 Super on a synthetic dataset modeled on nearly three decades of CRM deployments. Salesforce says it matches or exceeds leading models on CRM actions with three times fewer errors, and TechCrunch notes it uses fewer tokens for the same work. Pilots include Formula 1 and Xero, with general availability expected in winter 2026. The company that owns the customer record decided routine CRM work doesn’t need a frontier lab.
Oxford researchers proved a 38-year-old conjecture with GPT-6 Astra
Christian Coester, Elias Koutsoupias and Marek Zbysiński posted a proof of the k-server conjecture, posed in 1988 and repeatedly called the “holy grail” of competitive analysis. Their acknowledgments say who did what. The authors proved the three-server case “without AI assistance.” Discussions with ChatGPT 5.5 Pro and Gemini 3.1 Pro led to a more symmetric reformulation, and “ChatGPT 6 Astra subsequently derived an algebraic proof of correctness for any k.” The authors supplied a new representation, Astra adapted the proof to it and helped draft sections, and the authors revised the drafts. That’s a clearer account of AI’s share in a result than most announcements give.
Quick Hits
- Infrastructure: Temporal, whose software lets a crashed application pick up where it left off, closed $550 million at a $12.55 billion valuation. “As agents take on more critical work across more systems, every additional step creates another place to fail,” said CEO Samar Abbas.
- Safety: Bilal Chughtai, who left Google DeepMind’s AGI safety team in July, posted a warning this week that AI could kill us all and called for coordination “to avoid this manic race between AI companies.”
- Enforcement: Former FTC chair Lina Khan argued there are already plenty of laws to hold AI companies, and potentially their executives, accountable.
- Philanthropy: The Gates Foundation pledged $1 billion over two years to improve access to AI, and OpenAI committed $50 million to train health workers in Rwandan clinics.
- Agent audits: AIUC, from early Anthropic employee Rune Kvist and former METR COO Rajiv Dattani, raised a $40 million Series A led by Ribbit Capital to test agents against AIUC-1, a standard modeled on SOC 2.
- Chips: Samsung co-led a $230 million round for Dutch inference-chip startup Euclyd, and Cornelis raised $205 million led by IAG Capital Partners.
- Physical AI: Rome’s Exein raised $270 million at a $1.7 billion valuation, led by Headline.
- Hiring: Jack & Jill raised a $40 million Series A for two agents, Jack for candidates and Jill for employers, which have produced 25,000 interviews. The company doesn’t say how many became jobs.
- Legal: Morgan & Morgan committed at least $1 billion over ten years to legal tech and AI as it unveiled MX2, its own legal AI platform.
- Hacker News: “Dario, Please,” a takedown of Amodei’s essay, reached 579 points, and Andon Labs’ Pion, “an agent designed to run any company autonomously,” drew 547 comments.
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."
