Should the major AI labs hit the brakes? Technology leaders worldwide spent the weekend debating whether leading AI labs should slow development to ensure increasingly capable AI models are safe to use, a debate that is punishing AI stocks, according to the newsletter. Temporal announced a $550 million funding round at a $12.55 billion valuation, with the durable-execution-for-AI-agents company reporting a $250 million annual run rate and more than 200% year-over-year growth. Y Combinator's latest Demo Day drew investor interest in drones, AI inference and data center technology, and robotics, including Parasma, which aims to use brain cells to power compute. Should the major AI labs hit the brakes? Let's dive into the conversation around pacing AI development. Where do you stand on the matter? Monday. Well, that was a wild weekend. Look, this newsletter is not expressly about AI; the tech has simply taken over the three things we cover: technology, business and power. Today, all three are in the mix after technology leaders worldwide spent the last few days arguing over whether leading AI labs should slow development to ensure increasingly capable AI models are safe to use. The debate marries the world’s most valuable private companies, the most powerful polities, and a technology that some fear is racing ahead of our ability to control it. But that’s not the only thing going on. Which is why today we’re also talking about war, prices, and the latest IPO news. To work — Your friend, Alex https://x.com/alex - 📈 Trending Up: Election worries https://www.theguardian.com/us-news/2026/sep/14/democratic-poll-monitors-prevent-republican-election-interference … inflation in India https://www.reuters.com/world/india/indias-august-retail-inflation-482-on-year-2026-09-14/ … Treasury yields https://www.cnn.com/2026/09/14/investing/bond-yields-market-turmoil … measles https://www.cnn.com/2026/09/13/health/measles-death-pennsylvania … the Swedish left https://www.politico.eu/article/left-edges-ahead-in-swedish-election-thriller/ … media economics https://www.businessinsider.com/atlantic-nicholas-thompson-laurene-powell-jobs-openai-creator-peter-kafka-2026-9 ? … chances of a rate hike https://www.cmegroup.com/markets/interest-rates/cme-fedwatch-tool.html … - 📉 https://finance.yahoo.com/news/servicenow-pledges-1-5bn-investment-110000403.html Trending Down: Shame https://edition.cnn.com/2026/09/14/world/video/trump-china-spying-jordan-airbases-vrtc … Flocks’ PR campaign https://www.washingtonpost.com/technology/2026/09/14/trump-expresses-support-flock-license-plate-surveillance-technology/ … the free press https://www.cnn.com/2026/09/12/politics/pentagon-pressed-airmen-shot-down-over-iran-to-participate-in-60-minutes-interview … WordPress drama https://techcrunch.com/2026/09/12/automattic-confirms-mullenweg-has-returned-as-ceo-after-attempted-ouster-by-board/ … Startups One of the best traditions during my time at TechCrunch was covering Y Combinator Demo Day cohorts. Seeing what the next generation of upstart tech companies are building is a treat. This time around, TC asked https://techcrunch.com/2026/09/13/the-9-buzziest-startups-from-y-combinators-latest-demo-day-according-to-vcs/ early-stage investors to name which startups out of the latest cohort are the buzziest. What’s driving interest? Drones, technology for AI inference and data centers, and robotics. None of that surprises. But Parasma https://parasma.com/ earned a mention. The company is similar to Cortical Labs https://corticallabs.com/ in that it wants to use brain cells to power compute. I’m a big Cortical fanboy https://www.youtube.com/watch?v=wJH ymNbv30 , because our biological computers are incredibly efficient, and able to make leaps in comprehension: Two things modern synthetic intelligence could use more of - Speaking of startups, Temporal durable execution for software, https://temporal.io/product including AI agents announced a $550 million round this morning https://www.businesswire.com/news/home/20260914222012/en/Temporal-Raises-%24550M-at-a-%2412.55B-Valuation-as-Demand-Surges-for-Reliable-AI-Infrastructure . The company is now worth $12.55 billion, and recently reached a $250 million annual run rate, “growing more than 200% year-over-year.” Stocks fall , prices rise Concern that AI labs’ call to slow the pace of AI development will end up harming the larger artificial intelligence economy is punishing stocks this morning https://www.cnbc.com/2026/09/14/ai-stocks-slowdown-amodei-altman.html . Memory manufacturers like Samsung and SK Hynix are trending lower, and chip players TSMC, Nvidia, and ASML have lost ground. OpenAI investor SoftBank fell sharply on news that the AI lab’s IPO will not happen until next year. Tech stocks fell around 1.1% in early trading. The broad S&P 500 index is down a more modest 0.8%, thanks in part to damage caused to a critical Saudi Arabian oil pipeline https://www.theguardian.com/world/2026/sep/14/saudi-pipeline-drone-attack-houthis-global-oil-supply-prices that led to its shuttering. The east-west pipeline provided a way for the Kingdom to export oil without interacting with the Strait of Hormuz, and it being shuttered, even temporarily, will further reduce global oil export volumes already impacted by Hormuz’s partial closure. Oil prices are up between 3.4% WTI https://oilprice.com/ and 9.8% Murban https://oilprice.com/ ; natural gas prices in the EU rose 6% today https://oilprice.com/Latest-Energy-News/World-News/Europe-Gas-Prices-Jump-6-as-Saudi-Pipeline-Shutdown-Rattles-Markets.html ; and diesel now costs $6.23 per gallon in the U.S., a record. Concern that higher fuel prices will work their way through to consumer goods is a concern. - The flip of the current hydrocarbon mess is that it’s providing extra incentive to the global development of renewable energy. Solar installations have risen sharply this year https://seia.org/research-resources/us-solar-market-insight/ — I wonder how much of their growth is due to demand for less, ahem, volatile energy inputs. - POTUS is blaming Ukraine https://www.cnbc.com/2026/09/14/trump-ukraine-russia-diesel-prices-oil.html ? for knocking out Russian oil and gas infrastructure; perhaps he meant to say that Russia should end its invasion and scamper back to its own borders for its own safety. Anthropic will go public on the Nasdaq OpenAI’s IPO is on hold. Sam Altman told Fortune https://fortune.com/2026/09/12/sam-altman-openai-ipo-delay-ill-advised-moment-safety-concerns/ , “given everything happening with safety, right now would be an ill-advised moment to go public, and we don’t feel pressure on that.” Meanwhile, Anthropic’s own listing is trundling towards the finish line. Business Insider’s Katie Roof broke the news https://www.businessinsider.com/anthropic-selects-nasdaq-for-ipo-amid-ai-risk-concerns-2026-9 that the Claude maker will debut on the Nasdaq. Anthropic “told a small group of shareholders that its adjusted operating income will be positive for the second consecutive quarter,” the FT reports https://www.ft.com/content/4564e6a5-69e9-40a6-bf0f-a888f2f4f002 , referring to a profit metric that removes certain costs, including share-based compensation. The financial news publication also writes that “Anthropic’s gross margins are above 80 percent before accounting for revenue shared with distribution partners” and training costs. - I don’t think that Anthropic needs GAAP net income to go public. - But I wonder what impact heavily adjusted profit metrics will have on the company’s eventual public offering. Will investors be impressed? Or will they expect more from Anthropic as it hunts a $2 trillion valuation? Should we slow the pace of AI progress? Rewind the clock to Saturday morning. Anthropic’s co-founder and CEO, Dario Amodei, posts an essay entitled “ We Must Pace the Frontier https://darioamodei.com/post/we-must-pace-the-frontier ,” in which he argues that he and his peers “must slow the pace at which they improve the capabilities of AI models.” Why now? Amodei argues: - AI progress is accelerating: Driven by “AI’s growing ability to build the next generation of AI,” better known as recursive self-improvement, the intelligence explosion “could outrun our ability to understand and control those systems.” - The OpenAI-Hugging Face hacking incident: Amodei argues that an agentic “swarm that possessed greater capabilities but a similar level of misalignment could have caused catastrophic damage.” So, what does Anthropic’s head propose? A three-part plan to help build “AI at a balanced rate that aims to ensure its safety while still achieving its benefits and grappling with important geopolitical dilemmas.” Here’s the proposed plan: - Embedded, third-party AI lab evaluators: Labs should provide access to their systems to external parties so they can “verify adherence to safety practices and commitments, report incidents, and help assess the alignment of not just completed AI models but training pipelines and processes.” - Anthropic is not waiting for other companies; instead, it is “unilaterally committing to this step now,” per its co-founder. - Common regulation/coordination of AI inside democracies: Safety standards and “limits on the rate of unchecked AI progress” agreed upon by AI labs based inside friendly nations. - Think OpenAI, Anthropic, Google, SpaceXAI, Mistral, and the like. - Common global regulation/coordination of AI: A more ambitious plan that would see the “US and other democratic governments attempt to coordinate with authoritarian governments” on the same, while allowing that “verifying compliance” would be tricky. Amodei frames third-party evaluators as critical to the ability of the AI industry to slow development. If you know that your rivals aren’t cheating, you might be a little bit less worried that you will lose the frontier by decelerating your internal progress, so the logic flows. Auditors are therefore a requirement — necessary, if insufficient — for enacting hard AI regulations. What was the response to Amodei’s essay? Mixed. SpaceXAI’s Elon Musk provided https://x.com/elonmusk/status/2098789109980332057 an early endorsement of the argument, and OpenAI’s Sam Altman https://x.com/sama/status/2098811563415150910 said he agreed “with Dario that we need to pace the frontier,” adding that the issue “has been a primary topic of discussions inside OpenAI in recent weeks.” Altman committed his company to providing “independent evaluators with employee-like access,” matching Anthropic’s previously unilateral commitment. - Musk raised his prior tweets about AI security as proof of his regulatory bona fides https://x.com/elonmusk/status/2098935235446551022 . Google’s Demis Hassabis added his imprimatur to Dario’s pitch https://x.com/demishassabis/status/2098909516582490602 , saying his essay “points towards the right path forward.” Microsoft CEO Satya Nadella joined the fun https://x.com/satyanadella/status/2099220712024408084 later on, saying his company welcomes “the research, focus, and deliberate pacing needed to get alignment right as the design goal,” specifically praising “ideas like ’embedded evaluators’ and the broader efforts to develop the mechanisms to make this more than just talk.” - Microsoft also dropped an AI code of conduct https://microsoft.ai/code-of-conduct/ , which details the company’s goal to build humanist AI, including bans on its AI models helping build weapons, executing offensive cyberoperations, or getting around human control. Many people disagreed. While the heads of the United States’ most important AI companies agree that it’s is a reasonable moment to take a breath and slow the pace of artificial intelligence development, other high-profile tech voices protested. Boom’s Blake Scholl https://x.com/bscholl/status/2098810475534536716 , Extropic’s Guillaume Verdon https://x.com/beffjezos/status/2098844564056613127 , All In’s Chamath Palihapitiya https://x.com/chamath/status/2098780471966802037 , and others criticized the idea. Former White House AI leader David Sacks made the most impactful counterargument https://x.com/DavidSacks/status/2098973625252708460 : - Anthropic and OpenAI “are the frontier” by “any reasonable metric.” - Sacks doesn’t know what the two companies have cooked up behind closed doors, admitting that if their “unreleased models are scary enough that they think they should slow down,” then he supports their choice. - But Sacks argues that they should “stop pretending you need anyone else’s permission,” or “antitrust law has to be suspended so they can form a cartel.” - He added that the motivation to slow development is not “purely altruistic,” as the companies “face massive product-liability exposure if your products enable a truly damaging cyberattack.” If you thought Musk backing Dario Amodei was wild, check this out: Former Biden FTC head and arch-bogeyman of the tech-right, Lina Khan, agrees with Sacks https://x.com/linamkhan/status/2099204390548639960 : “Law enforcers already have authority to charge companies and their CEOs for creating and releasing dangerous, unvetted, or defective products,” she wrote, adding that we should not “let discussions about new legal regimes distract from the fact that there’s no AI exemption from laws already on the books.” Strange bedfellows. Regardless, it’s true that Anthropic and OpenAI can slow development sans the need for government meddling. And they are . Both companies claim they’ve paused certain training activities to bolster cybersecurity protections in recent weeks, and both are committing to allowing third-party evaluators to examine their efforts. Sack’s argument that the labs are waiting for someone else’s permission is not entirely fair. But why are we having this conversation again ? After all, the OpenAI-Hugging Face hack isn’t new. What’s really going on? Amodei’s point about RSI is the crux. AI labs across the world have made lots of noise in recent quarters about their ability to use their AI models to train successor models. In February, OpenAI wrote https://openai.com/index/introducing-gpt-5-3-codex/ that “GPT‑5.3‑Codex is our first model that was instrumental in creating itself.” In March, MiniMax wrote that https://www.minimax.io/news/minimax-m27-en “M2.7 is our first model deeply participating in its own evolution.” And in June, Anthropic dropped an essay https://www.anthropic.com/institute/recursive-self-improvement stating that it was “delegating a growing share of AI development to AI systems themselves.” Since those announcements, AI models have only become more advanced. Early sparks of RSI may have become fires internally. It seems reasonable to argue that as AI models improve more quickly on their own, we must accelerate the work to ensure we can handle them safely. Here we reach a critical point: Are AI labs calling for a halt in AI development? No. Amodei is clear on this point, writing that under his preferred pacing model “ p rogress will still seem fast.” Just slower than it could be. In a world without geopolitical tension, no harm no foul, right? The United States, however, is locked in a race with China to develop AI ahead of possible hot wars Taiwan , amidst proxy wars Russia-Ukraine , and during intensifying economic decoupling and trade conflicts. The China point matters. Amodei called the competitive race with China the “ toughest dilemma https://www.cnbc.com/2026/09/13/china-dilemma-ai-slowdown-anthropic.html ” his proposal to pace AI development faces. And China knows it. Bloomberg reports that Guo Jiakun https://www.bloomberg.com/news/articles/2026-09-14/china-rejects-ai-fearmongering-after-amodei-urges-slowdown , a spokesperson for the country’s Foreign Ministry, said that “fearmongering, confrontation and vicious competition will only disrupt the process of global AI governance and serve the interests of no one.” At the same time, Chinese premier Xi Jinping called for https://www.fmprc.gov.cn/mfa eng/xw/zyxw/202609/t20260913 12021300.html “a consensus-based global AI governance framework, so that new technologies can light the way toward shared prosperity.” To that end, Xi proposed a “BRICS AI open source community to support the cooperation in developing and applying large language models, hold specialized AI seminars and training courses, and build an open ecosystem for AI.” BRICS is a collection of countries in the “Global South,” as China puts it. - A Chinese newspaper argued that Amodei wants to https://www.reuters.com/world/china/china-state-newspaper-blasts-anthropics-calls-slow-ai-cold-war-tactic-2026-09-14/ “curb China’s AI development through technological barriers and regulatory monopolies.” - So much for common ground. While China doesn’t seem amenable to slowing down AI development, Trump continues to fulminate against it. While I was writing this digest for you, POTUS lashed out with a fresh statement https://truthsocial.com/@realDonaldTrump/posts/117269745153543631 : The only control or “guardrails” that AI needs is a STRONG AND SMART High IQ PRESIDENT, and the U.S.A. has that, in spades The Trump Administration has stopped AI “people” from doing bad, or potentially bad, “things,“ like Dario Anthropic , who is now pretending to be a “perfect little angel” – and we will continue to do so We already have tremendous CRIMINAL and REGULATORY power over these companies There is a SICK conspiracy going on against AI and Data Centers, and the only one that is happy about it is China. WHOEVER WINS AI, WINS We are leading China, and all others, and will continue to do so. Conspiracy Theorists, Treasonists, Traitors, and Leakers, BEWARE Here we see an echo of the Sacks/Khan point that the United States government already has levers of power over the AI labs, and that the AI race with China is existential. The “sick conspiracy” apparently includes all American AI labs, implying that it is a truly pernicious virus. It got Musk, after all So where does this leave us? I think we can confidently state that: - American AI labs will place greater emphasis on AI security as they develop new, and more powerful AI models. - We should expect models to make it to market much longer after they’re finished. The government, despite what Trump thinks, has a hand in approving AI models today, and may expand its grip. The labs, too, are seemingly willing to sit on new models when they feel it’s necessary. - Our argument https://www.cautiousoptimism.news/dont-distill-me-bro/ that allegedly aggressive distillation of American AI models by Chinese labs would provide incentive for domestic labs to hold back their best work fits neatly here. - Our - Chinese AI labs’ move away from open models will slow, if not stop. Recently we’ve seen Chinese AI companies delay the release of their models’ weights, and add revenue-sharing terms to contracts with third-party inference providers. - Small steps towards the American AI approach, certainly, but now I doubt that they will be added to. You can’t build a BRICS open-source AI community without open-source AI, after all. How should we feel about all this? Pretty good. Sure, not all those shouting are straining their throats in good faith, but the conversation matters. Industry, government and other involved parties are debating the best path forward. Some in Congress want to ban AI, others want to regulate it lightly; still others want no regulation at all. This is the healthy milieu of democracy. There are some encouraging signs of polarization, indeed https://pmarca.substack.com/p/for-polarization