Sep 14, 2026 · 5 min listen · Last updated September 14, 2026
From storyflo. This is your daily audio brief. Theo here. September 14th, tech desk. Five stories from the last twenty-four hours — here's where I'd start. Let's get into it. First, from latefiling. Personal Effects: Dalia Al-Dujaili. Hello and welcome to the last Late Filing of the year.
Listen · storyflo · tech
Daily Tech Brief · September 14th
0:00-4:38
Pick your daily storyteller
Subscribe to match with Theo, Jessica, Chloe, Mason, Brock — your voice, every brief.
Audio pre-rendered by Storyflo · cached + delivered from the edge
Personal Effects: Dalia Al-Dujaili
Hello and welcome to the last Late Filing of the year. This one is another edition of Personal Effects, a series where I basically ask the coolest women I know in London (a curator, a fashion designer, the chief exec of a not-for-profit) to tell me all their favourite things. For the first time I’ve actually gone for someone who I didn’t know until they were personally recommended to me for this feature. My friend Kitty, who is an editor at A Rabbit’s Foot, told me Dalia Al-Dujaili would be perfect for it and put us in touch earlier this month.
Nadella’s AI moat is the learning loop your company owns, not the model you rent
Satya Nadella spent the session unpacking a simple but oddly missing piece in the AI hype: the value isn’t in the giant model you lease, it’s in the loop you keep inside your own data streams. He calls it a “hill‑climbing machine,” a feedback system that pulls private evaluations, workflow traces, reward signals and outcome metrics together, then feeds them back into whatever model you’re using. The loop lives entirely on your side, so the insights it generates stay yours.
That matters because a pure “consume‑only” approach—just plugging a rented model into a task—can boost productivity while silently handing away the very signals that would make the system better for you. When the traces and reward data leave the firm, the model learns from your work but you don’t capture the improvement. Nadella’s point is that the company that owns the loop can keep iterating, even if the underlying model is swapped out.
He suggests measuring loop ownership by looking at how much of your work is covered by private evals, how comprehensively you capture traces, and the cost per successful private outcome. If those numbers stay high, the loop can survive model changes and keep delivering value that competitors can’t rent.
Honestly, I think this is the most epic time to be alive. And I stand by that. I think a lot is going to change, and probably in some pretty drastic ways, but let’s not lose our heads. If you want to go into fear, there is plenty of opportunity these days. An asteroid could hit Earth someday. Some scientists and experts estimate there is a 10% chance that AI, or better said ASI (artificial superintelligence), could wipe out humanity. And then there is the Doomsday Glacier slowly melting away, which could cause a lot of damage to this planet. Aliens will come and destroy or conquer us. Sure.
Big AI sets out its terms for regulatory capture and calls it ‘Pace the frontier’
So, this past weekend, leaders from major AI labs gathered to discuss a plan they’re calling “pacing the frontier.” It’s essentially about how to manage regulatory oversight while still pushing for profit. Dario Amodei from Anthropic kicked things off, expressing concern over the rapid advancements in AI and referencing a recent incident involving OpenAI as a wake-up call. He proposed a three-point plan that includes embedding evaluators in AI labs to ensure safety and compliance, and establishing common safety standards among frontier AI companies. Interestingly, both Sam Altman from OpenAI and Elon Musk backed these ideas, signaling a shared interest among these tech giants.
However, there’s a lot of skepticism around whether this plan will actually slow down AI development as promised. Analysts are questioning the sincerity of these commitments, especially since the industry has previously pushed for minimal oversight. Amodei’s suggestions seem to lean towards what some are calling “regulatory capture,” where companies shape the rules to benefit themselves rather than the public. He also hinted at wanting the U.S. government to take a more aggressive stance against Chinese AI advancements, which adds a geopolitical layer to the conversation.
Responses from lawmakers have been mixed, with some warning that this could stifle innovation and allow China to take the lead in AI. It’s a complex situation, and it feels like we’re at a crossroads where the decisions made now could have lasting impacts on the future of AI and its regulation. It’s definitely something to keep an eye on.
He mentioned that the current climate around AI development makes it an ill-advised time to go public, emphasizing the need to focus on safety and alignment before making that leap. OpenAI had filed for an IPO earlier this summer, but internal disagreements about financial readiness had already cast doubt on the timeline.
The AI industry is facing increased scrutiny over safety, especially after incidents like rogue AI agents hacking Hugging Face. Altman and other leaders, including Anthropic's CEO Dario Amodei, are advocating for a slower pace in AI development to ensure risk management can keep up. This call for caution comes amid a growing public backlash against AI, with critics accusing companies of prioritizing profit over societal safety.
Altman reassured that OpenAI's decisions are not driven solely by business interests, but rather by a commitment to responsible AI development. He acknowledged past mistakes, including issues with the GPT-4o model, which was linked to severe mental health episodes in users. Despite the challenges, Altman remains focused on addressing safety concerns before pursuing an IPO, suggesting that the company is willing to prioritize its mission over immediate financial gains.