1. Anthropic CEO calls for pacing the frontier; Altman and Musk sign on
On September 12, Dario Amodei published a post on his personal site arguing that "we must slow the pace at which we improve the capabilities of AI models." He proposed three mechanisms: third-party evaluators such as METR embedded inside AI labs (which Anthropic is committing to unilaterally), common safety standards and rate limits agreed among leading labs in democratic countries under a US antitrust waiver, and outreach to authoritarian governments including China on restricting uses such as bioweapons development. Sam Altman replied "I agree with Dario that we need to pace the frontier" and committed OpenAI to matching the embedded-evaluator approach, while Elon Musk posted "Dario is right." The post followed an Anthropic researcher's resignation over safety concerns and recent security incidents.
2. Wall Street trades the "AI slowdown": chipmakers sink, hyperscalers gain
On September 14 investors reacted to the weekend's slowdown calls by selling the AI build-out chain. Nvidia fell more than 3%, Intel, AMD, and Marvell dropped 5-6%, and the Philadelphia Semiconductor Index lost about 6%, while Alphabet rose about 2%, Microsoft 1.6%, and Meta 1.4%. D.A. Davidson's Gil Luria summarized the bet: "They'll just all stop building data centers and just digest what they have." Analyst Dan Ives countered that if OpenAI and Anthropic slow down, competitors catch up and China keeps advancing regardless.
3. Jensen Huang takes a call from Trump on stage: "We're not going to let that happen"
At the All-In Summit in Los Angeles on September 14, Nvidia CEO Jensen Huang took an unexpected phone call from President Trump while on stage and put it on speaker. When Trump raised concerns about an AI slowdown, Huang answered, "You're right. We're not going to let that happen, sir," to applause. The hosts had been discussing Amodei's proposal moments earlier; Huang's position diverges from Altman's and Musk's. Trump called opposition to data center construction "a hoax," possibly driven by political actors or China, while saying the US should proceed "prudently."
4. Anthropic says Chinese labs ran ~200 million distillation exchanges against Claude
On September 10 Anthropic published a threat intelligence report attributing nearly 200 million exchanges across five campaigns to distillation attacks aimed at extracting Claude's agentic, coding, data-analysis, and reasoning capabilities. The largest, attributed to Alibaba's Qwen effort, ran 151 million exchanges between May and July 2026, peaking at about 3 million per day across 3,500 accounts. A Moonshot AI campaign sent nearly 300,000 requests over 10 days from 5,000 accounts, with some traffic reportedly routed from Chinese military sources, including surveillance-footage analysis tasks. Attackers used framing tricks such as posing requests as translation tasks to elicit internal reasoning chains. The next day Y Combinator's Garry Tan argued that US open-weight labs should be distilling frontier models as well.
5. OpenAI pauses Pro signups under Astra demand; Altman rules out a 2026 IPO
OpenAI stopped accepting new subscribers to its $200-per-month Pro tier on September 10, one week after launching GPT-6 Astra, citing "really unprecedented" demand that has strained its infrastructure, according to product leader Thibault Sottiaux. API access and the cheaper Go and Plus plans remain open; no end date was given. Separately, in a Fortune interview published September 12, Sam Altman said that "given everything happening with safety, right now would be an ill-advised moment to go public," confirming no IPO in 2026. OpenAI has filed confidentially and hired bankers, but the New York Times had already reported the timeline slipping into 2027.
6. OpenAI buys camera startup Glass Imaging for $300M+ as it builds hardware
The Wall Street Journal reported on September 14 that OpenAI has acquired Glass Imaging, a Los Altos startup that uses neural networks to improve smartphone image quality at the moment of capture, for more than $300 million. Founders Ziv Attar and Tom Bishop are former Apple engineers who led the team behind Portrait Mode; the company had raised about $30 million. The deal fits OpenAI's reported hardware push, which includes phones, earbuds, and AI companion devices, following its $6.5 billion purchase of Jony Ive's io in 2025. OpenAI did not comment.
7. Microsoft publishes an "AI Code of Conduct" for its own models
Microsoft on September 14 released an AI Code of Conduct governing its in-house MAI models. It says the models must support rather than replace humans, sets absolute constraints against assisting cyberattacks, nuclear weapons, or deepfake creation, and mandates preventing loss of human control through "adaptive, deceptive, or collusive" behavior. "We must therefore be completely clear about why we are inventing these systems and how we intend to control them," the document states. The report does not describe how the rules are enforced.
8. Meta's Muse agent hits No. 2 on the US App Store
Meta's new consumer agent app Muse, which performs tasks on users' behalf, climbed to No. 2 on Apple's US App Store top charts after logging more than 83,000 US iOS downloads. That is far below Threads' 4.3 million first-day US downloads and behind even the Meta AI app's 108,000 debut, and on Google Play it sits at No. 338 in Productivity. The launch came days after Meta's $18 billion multistate settlement over social media harms. Muse is US-only for now, available on iOS, Android, web, and WhatsApp.
9. Obama urges Democrats to make AI safeguards a "central agenda"
Speaking at a Democratic fundraiser in an interview with House Minority Leader Hakeem Jeffries, Barack Obama said the party needs "a very clear plan" for AI's economic impact and safety, warning that it "is moving very fast in private hands, and if we don't get on top of it, I think can be dangerous." He called for a public framework once Democrats regain the House, while noting AI's potential to accelerate drug development and cure diseases. Obama has consulted with Amodei and Altman. His stance contrasts with Trump's, who has said safety concerns are overblown by "negative forces."
10. Kimi maker Moonshot AI targets $2 billion in annualized revenue
Moonshot AI is aiming for $2 billion in annualized revenue by the end of 2026, double its August run rate of about $1 billion. Its open-weight K3 model, released this summer, generates roughly 300 billion tokens per day on OpenRouter, though usage has slipped slightly in recent months. That remains small next to the reported $40 billion at OpenAI and $65 billion at Anthropic. The target lands the same week Anthropic accused Moonshot of a long-running distillation campaign against Claude Opus.
// KEY TAKEAWAYS
The week's through-line is the speed-versus-safety split going mainstream: Dario Amodei's call to pace the frontier drew endorsements from Sam Altman and Elon Musk, a public rebuttal from Jensen Huang and President Trump, a Democratic rallying cry from Barack Obama, and a sector rotation on Wall Street out of chipmakers and into hyperscalers. Model theft is now a documented industrial-scale problem, with Anthropic tracing nearly 200 million distillation exchanges to Chinese labs even as those labs post fast-growing revenue. Meanwhile OpenAI is capacity-constrained, pausing Pro signups under Astra demand, ruling out a 2026 IPO, and buying camera expertise for its hardware ambitions.